DOWNLOAD MY Ph.D Thesis - UNAM

I would like to thank all the people who have influenced me during my postgraduatestudies. To all of you, thank you very much. I would like to extend a very special thanksto a number of people that have made this dissertation possible.I especially wish to thank the ‘Universidad acional Autónoma de México’ (UNAM)for sponsoring my research at the University of Sheffield.I am indebted to my academic supervisor Dr. Rob Edwards from the EEE Department,at University of Sheffield for his guidance and support during the last two years of myresearch. He taught me how to do research without losing the practical prospective.I would also thank my Mexican tutor Dr. Salvador Landeros Ayala and Mtra. ArtemisaPedroza de Gortari from the Facultad de Ingenieria, UNAM for helping me to renewmy scholarship several times at the UNAM.A very special thanks to my first academic supervisor Prof. Colin Smythe from theComputer Science Department, at the University of Sheffield for supporting me andproviding the stimulation research environment in which this work has been carried out.I also thank him for accepting to review this dissertation.I acknowledge my tutor Prof. Graham Rees and my second supervisor Dr. MohammedBenaissa for providing general comments about my research.I need to specially thank Dr. Klaus-Dieter Schunke from the Institute forCommunications Technology, Braunschweig Technical University, Germany forproviding me with technical information for this research and his careful reviews ofsome of my papers.I would like also to thank the students from the Centre For Mobile CommunicationsResearch (C4MCR) group, University of Sheffield, who have made this a pleasant time.My special gratitude to my parents and brothers for their great help in backing me uppsychologically in Mexico throughout the period of this project.vi

LIST OF PUBLICATIOSJournal Papers(1) V. Rangel, R. Edwards, P. Tzerefos and K. Schunke. “Delivery of Low Bit Rate IsochronousStreams over the Digital Video Broadcasting/Digital Audio-Video Council internationalcable television protocol”, Accepted for publication in May 2002, IEEE Tran. onBroadcasting, Ref. No. BTS-00-68.(2) V. Rangel, R. Edwards, and K. Schunke, “Contention Resolution Algorithms for CATVNetworks Based on the DVB/DAVIC Cable Modem Protocol Specification (ETS EN 200800)”, Accepted for publication, Cable Telecommunication Engineering (CTE), Sep. 2002.(3) V. Rangel and R. Edwards, “Performance Evaluation of CATV Networks Based on theEuropean Cable Communications Protocol: DVB/DAVIC”, Submitted for publication, IEEETran. on Broadcasting, Jun. 2002.(4) V. Rangel and R. Edwards, “Performance Analysis of QoS over CATV Networks Based onthe DVB/DAVIC Protocol”, Submitted for publication, CTE, Jun. 2002.(5) V. Rangel and R. Edwards. “Performance Evaluation of Adaptive Contention Slot Allocatorsfor CATV Networks based on the European Cable Communications Protocol”. Submitted toIEEE GLOBECOM-2002, Mar. 2002.Conference Papers(6) C. Smythe, P. Tzerefos, V. Sdralia, V. Rangel and S. Cvetkovic, “Performance Evaluationof the DVB/DAVIC Cable Return Channel Path for Interactive Services”, IBC DVB'99,Apr. 1999.(7) C. Smythe, P. Tzerefos, V. Sdralia, V. Rangel and S. Cvetkovic, “Choosing the DOCSIS orDVB/DAVIC Return Channel Path for Interactive Services”, IBC Digital InteractiveRetailing '99 Conference, Oct. 1999.(8) C. Smythe, P. Tzerefos, V. Sdralia, V. Rangel and S. Cvetkovic, “Cable Modems and theReturn Channel Path for Interactive Services: DOCSIS vs. DVB - PerformanceEvaluation”, IBC Television Distribution '99 Conference, May 1999.(9) V. Rangel, C. Smythe, P. Tzerefos, S. Cvetkovic and S. Landeros, “A comparison of theDOCSIS, DVB/DAVIC and IEEE 802.14 Cable Modem Specifications”, Proc. of theInternational Conference on Telecommunications (ICT 2000), ISBN 968-36-7763-0,Acapulco, May 2000.(10) V. Rangel, R. Edwards, and K. Schunke, “Contention Resolution Algorithms for CATVNetworks Based on the DVB/DAVIC Cable Modem Protocol Specification (ETS EN 200800)”, Proc. of the International Broadcasting Conference (IBC), Amsterdam, Sep. 2001.(11) V. Rangel and R. Edwards. “Performance Analysis and Optimisation of the Digital VideoBroadcasting/Digital Audio Visual Council Cable Modem Protocol for the Delivery ofIsochronous Streams”. Proc. of GLOBECOM-2001, IEEE, ISBN 0-7803-7206-9, Vol. 1,pp. 430-434, Nov. 2001.vii

CHAPTER 6: OPTIMISATIO OF CRA ALGORITHMSUSIG ADAPTIVE CSA 6-16.1 INTRODUCTION 6-16.2 CONTENTION SLOT ALLOCATOR 6-26.2.1 HOW MANY CONTENTION-SLOTS PER SIGNALLING FRAME? 6-26.2.2 SIMPLE-CSA: PERFORMANCE OPTIMISATION OF THE USE OF CONTENTION SLOTS 6-36.2.2.1 Exponential backoff algorithm performance using a Simple-CSA6.2.2.2 Splitting tree algorithm performance using a Simple-CSA6-36-66.2.3 ENHANCED - CSAS 6-96.2.3.1 Forced-CSA used with the exponential backoff algorithm 6-106.2.3.2 Variable-CSA used with the exponential backoff algorithm 6-116.2.3.3 Performance comparison of Enhanced-CSAs 6-126.3 DYNAMICS OF THE DVB CONTENTION RESOLUTION ALGORITHMS 6-176.3.1 EXPONENTIAL BACKOFF ALGORITHM OPTIMISATION 6-176.3.2 SPLITTING TREE ALGORITHM OPTIMISATION 6-206.3.3 PERFORMANCE COMPARISON BETWEEN THE EXPONENTIAL BACKOFFALGORITHM AND THE SPLITTING TREE ALGORITHM 6-236.4 CONCLUSIONS 6-29CHAPTER 7: PERFORMACE OPTIMISATIO FOR THE SUPPORTOF TCIS AD A PERFORMACE COMPARISO OFDVB/DAVIC AD DOCSIS 7-17.1 INTRODUCTION 7-17.2 ENHANCED-RESERVATION-REQUEST MECHANISMS 7-27.2.1 RESERVED REQUEST (RR) 7-27.2.2 CONTINUOUS RESERVED REQUEST (CRR) 7-37.2.3 ENHANCED PURE RESERVATION ACCESS (EPRA) 7-37.2.4 PIGGYBACK REQUEST (PG) 7-47.2.5 CONTINUOUS PIGGYBACK REQUEST (CPG) 7-57.2.6 UNSOLICITED GRANT SLOT (UGS) 7-67.2.7 PERFORMANCE COMPARISON OF ENHANCED-RESERVATION-REQUESTMECHANISMS 7-77.3 QUALITY OF SERVICE (QOS) 7-137.3.1 PRIORITISATION IN QOS 7-147.3.2 RESERVATION (FIXED-RATE ACCESS) IN QOS 7-177.4 COMPARISON: DVB/DAVIC VS. DOCSIS 7-247.4.1 TECHNICAL COMPARISONS 7-257.4.1.1 MAC layer comparisons 7-257.4.1.2 PHY layer comparison 7-267.4.2 PERFORMANCE COMPARISONS 7-277.5 CONCLUSIONS 7-36x

LIST OF FIGURESFIGURE 2.1 – EXAMPLE BLOCK DIAGRAM OF A TYPICAL CATV NETWORK. 2-3FIGURE 2.2 – SPECTRUM ALLOCATION ON A CATV NETWORK. 2-4FIGURE 2.3 – MODERN CATV NETWORK WITH HFC ARCHITECTURE. 2-6FIGURE 2.4 – CABLE MODEM CONFIGURATIONS. 2-7FIGURE 3.1 – DVB SYSTEM REFERENCE MODEL. 3-3FIGURE 3.2 – DVB/DAVIC INITIALISATION AND REGISTRATION PROCESS. 3-5FIGURE 3.3 – CONTENTION-RESOLUTION-GRANT CYCLE. 3-7FIGURE 3.4 – BLOCK DIAGRAM OF THE EXPOETIAL BACKOFF ALGORITHM. 3-8FIGURE 3.5 – BLOCK DIAGRAM OF THE SPLITTIG TREE ALGORITHM. 3-10FIGURE 3.6 – SLOT STRUCTURE. 3-11FIGURE 4.1 – STAGE DIAGRAM OF OPNET SIMULATION MODELS. 4-3FIGURE 4.2 – GENERIC CSF MODEL. 4-5FIGURE 4.3 – NETWORK TOPOLOGY. 4-7FIGURE 4.4 – NIU NODE DESCRIPTION. 4-8FIGURE 4.5 – INTERNET PACKET DISTRIBUTION. 4-8FIGURE 4.6 – NIU MANAGER PROCESS – FSM. 4-9FIGURE 4.7 – NIU CHILD PROCESS – FSM. 4-10FIGURE 4.8 – PROTO-C CODE OF THE STATE: TX. OPPORTUNITY. 4-11FIGURE 4.9 – INA NODE DESCRIPTION. 4-12FIGURE 4.10 – INA MANAGER PROCESS – FSM. 4-13FIGURE 4.11 – INA TX CONTROLLER - FSM. 4-14FIGURE 4.12 – UPSTREAM CHANNEL MODEL WITH VIRTUAL CONTENTION RESERVATION ANDFIXED-RATE CHANNELS. 4-15FIGURE 4.13 – UPSTREAM TIMING DIAGRAM. 4-19FIGURE 4.14 – MEAN SERVICE TIME WHEN THE SYSTEM IS IN IDLE STATE. 4-22FIGURE 4.15 – SCHEDULER LOOK AHEAD DELAY. 4-23FIGURE 4.16 – VARIANCE OF THE MEAN SERVICE TIME. 4-28FIGURE 4.17 – TIME OF BUSY CYCLE. 4-28FIGURE 4.18 – SYSTEM THROUGHPUT: SIMULATION AND THEORETICAL RESULTSFOR DIFFERENT PACKET SIZES. 4-32FIGURE 4.19 – MEAN ACCESS DELAY: SIMULATION AND THEORETICAL RESULTSFOR DIFFERENT PACKET SIZES. 4-33FIGURE 5.1 - MEAN ACCESS DELAY, SYSTEM THROUGHPUT AND UTILISATION VS. OFFERED LOAD. 5-10FIGURE 5.2 - CUMULATIVE PROBABILITY VS. MEAN ACCESS DELAY. 5-11FIGURE 5.3 - SYSTEM THROUGHPUT AND UTILISATION VS. NO. OF ACTIVE STATIONS, EXPOETIALBACKOFF AND SPLITTIG TREE ALGORITHM. 5-12FIGURE 5.4 – MEAN ACCESS DELAY VS. NO. OF ACTIVE STATIONS, EXPOETIAL BACKOFF ANDSPLITTIG TREE ALGORITHM. 5-13FIGURE 5.5 – MEAN ACCESS DELAY VS. NO. OF ACTIVE STATIONS FOR DIFFERENT CONTENTION MESSAGES.5-15FIGURE 5.6 – SYSTEM THROUGHPUT VS. NO. OF ACTIVE STATIONS FOR DIFFERENT CONTENTIONMESSAGES. 5-16FIGURE 5.7 – DATA TRANSMITTED IN THE CONTENTION-BASED REGION VS. NO. OF ACTIVESTATIONS FOR DIFFERENT CONTENTION MESSAGES. 5-17FIGURE 5.8 – BANDWIDTH CONSUMED BY COLLISIONS VS. NO. OF ACTIVE STATIONS FORDIFFERENT CONTENTION MESSAGES. 5-17FIGURE 5.9 – MEAN ACCESS DELAY VS. NO. OF ACTIVE STATIONS FOR DIFFERENT REQUEST SIZES. 5-19FIGURE 5.10 – SYSTEM THROUGHPUT VS. NO. OF ACTIVE STATIONS FOR DIFFERENT REQUEST SIZES. 5-20FIGURE 5.11 – RESERVATION REQUEST BANDWIDTH CONSUMED VS. NO. OF ACTIVE STATIONS. 5-21FIGURE 5.12 – MEAN ACCESS DELAY VS. NO. OF ACTIVE STATIONS FOR VOIP AND IP TRAFFIC. 5-22FIGURE 5.13 – MEAN ACCESS DELAY VS. NO. OF ACTIVE STATIONS FOR INCREASED BUFFER SIZE. 5-24FIGURE 5.14 – BUFFER CAPACITY VS. OFFERED LOAD FOR INCREASED BUFFER SIZE. 5-25FIGURE 5.15 – DISCARDED PACKETS VS. OFFERED LOAD FOR INCREASED BUFFER SIZE. 5-26FIGURE 5.16 – MEAN ACCESS DELAY VS. NO. OF ACTIVE STATIONS FOR DIFFERENT MCI VALUES. 5-28xii

FIGURE 5.17 – SYSTEM THROUGHPUT VS. NO. OF ACTIVE STATIONS FOR DIFFERENT MCI VALUES. 5-29FIGURE 5.18 – SYSTEM THROUGHPUT/UTILISATION VS. PACKET SIZE. 5-30FIGURE 5.19 – SYSTEM THROUGHPUT IN KBPS AND PACKETS/S VS. PACKET SIZE. 5-31FIGURE 5.20 – SYSTEM THROUGHPUT VS. NO. OF ACTIVE STATIONS FOR INCREASED PACKET SIZE. 5-32FIGURE 5.21 – MEAN ACCESS DELAY VS. NO. OF ACTIVE STATIONS FOR INCREASED PACKET SIZE. 5-33FIGURE 5.22 – BANDWIDTH CONSUMED BY COLLISIONS VS. NO. OF ACTIVE STATIONSFOR INCREASED PACKET SIZE. 5-34FIGURE 6.1 – MEAN ACCESS DELAY VS. NO. OF ACTIVE STATIONS.EXPOETIALBACKOFF ALGORITHM WITH A SIMPLE-CSA. 6-4FIGURE 6.2 – SYSTEM THROUGHPUT VS. NO. OF ACTIVE STATIONS. EXPOETIAL BACKOFF ALGORITHMWITH A SIMPLE-CSA. 6-4FIGURE 6.3 – AVERAGE NO. OF CSS PER RESERVATION REQUEST VS. NO. OF ACTIVE STATIONS.EXPOETIAL BACKOFF ALGORITHM WITH A SIMPLE-CSA. 6-5FIGURE 6.4 – MEAN ACCESS DELAY VS. NO. OF ACTIVE STATIONS. SPLITTIG TREE ALGORITHM WITHA SIMPLE-CSA. 6-6FIGURE 6.5 – SYSTEM THROUGHPUT VS. NO. OF ACTIVE STATIONS. SPLITTIG TREE ALGORITHM WITHA SIMPLE-CSA. 6-7FIGURE 6.6 – AVERAGE NO. OF CSS PER RESERVATION REQUEST VS. ACTIVE STATIONS. SPLITTIGTREE ALGORITHM WITH A SIMPLE-CSA. 6-8FIGURE 6.7 – SYSTEM THROUGHPUT VS. NO. OF ACTIVE STATIONS. EXPOETIAL BACKOFFALGORITHM WITH VARYING CSA APPLIED. 6-14FIGURE 6.8 – AVERAGE NO. OF CSS PER RESERVATION REQUEST VS. NO. OF ACTIVE STATIONS.EXPOETIAL BACKOFF ALGORITHM WITH VARYING CSA APPLIED. 6-15FIGURE 6. 9 – MEAN ACCESS DELAY VS. NO. OF ACTIVE STATIONS. EXPOETIAL BACKOFF ALGORITHMWITH VARYING CSA APPLIED. 6-15FIGURE 6.10 – CUMULATIVE PROBABILITY VS. NO. OF ACTIVE STATIONS.EXPOETIALBACKOFF ALGORITHM WITH VARYING CSA APPLIED.OFFEREDLOAD = 46% PRODUCED BY 34 STATIONS. 6-16FIGURE 6.11 – CUMULATIVE PROBABILITY VS. NO. OF ACTIVE STATIONS FOR DIFFERENT BACKOFFWINDOWS AND OFFERED LOAD OF 46% AND MIXED TRAFFIC. 6-19FIGURE 6.12 – MEAN ACCESS DELAY VS. NO. OF ACTIVE STATIONS FOR DIFFERENTBACKOFFWINWOWS AND VOIP TRAFFIC. 6-19FIGURE 6.13 – SYSTEM THROUGHPUT VS. NO. OF ACTIVE STATIONS FOR DIFFERENTBACKOFF WINWOWS AND VOIP TRAFFIC. 6-20FIGURE 6.14 – MEAN ACCESS DELAY VS. NO OF ACTIVE STATIONS FOR DIFFERENT ENTRYSPREADING FACTORS AND MIXED TRAFFIC. 6-21FIGURE 6.15 – MEAN ACCESS DELAY VS. NO. OF ACTIVE STATIONS FOR DIFFERENT ENTRYSPREADING FACTORS AND VOIP TRAFFIC. 6-22FIGURE 6.16 – SYSTEM THROUGHPUT VS. NO. OF ACTIVE STATIONS FOR DIFFERENT ENTRYSPREADING FACTORS AND VOIP TRAFFIC. 6-23FIGURE 6.17 – SYSTEM THROUGHPUT. EXPOETIAL BACKOFF VS. SPLITTIG TREE ALGORITHMFOR DIFFERENT TRAFFIC CONFIGURATIONS. 6-24FIGURE 6.18 – MEAN ACCESS DELAY. EXPOETIAL BACKOFF VS. SPLITTIG TREE ALGORITHMFORDIFFERENT TRAFFIC CONFIGURATIONS. 6-25FIGURE 6.19 – COMPONENTS OF DELAY, EXPOETIAL BACKOFF VS. SPLITTIG TREE ALGORITHM 6-26FIGURE 6.20 – UPSTREAM SLOT USAGE. EXPOETIAL BACKOFF VS. SPLITTIG TREE ALGORITHM 6-27FIGURE 7.1 – FUNCTIONALITY OF THE RR MECHANISM FOR DIFFERENT PACKET SIZES IN THETRANSMISSION QUEUE. 7-2FIGURE 7.2 – FUNCTIONALITY OF THE CRR MECHANISM FOR DIFFERENT PACKET SIZES IN THETRANSMISSION QUEUE. 7-3FIGURE 7.3 – FUNCTIONALITY OF THE EPRA MECHANISM FOR DIFFERENT PACKET SIZES IN THETRANSMISSION QUEUE. 7-4FIGURE 7.4 – DATA SLOT STRUCTURE WITH A PIGGYBACK REQUEST. 7-4FIGURE 7.5 – FUNCTIONALITY OF THE PG MECHANISM FOR DIFFERENT PACKET SIZES IN THETRANSMISSION QUEUE. 7-5FIGURE 7.6 – FUNCTIONALITY OF THE CPG MECHANISM FOR DIFFERENT PACKET SIZES IN THETRANSMISSION QUEUE. 7-5xiii

FIGURE 7.7 – MEAN ACCESS DELAY VS. NO. OF ACTIVE STATIONS FOR VOIP AND MIXED TRAFFIC.EXPOETIAL BACKOFF ALGORITHM WITH EHACED-RESERVATIO-REQUEST MECHAISMS. 7-8FIGURE 7.8 – SYSTEM THROUGHPUT VS. NO. OF ACTIVE STATIONS FOR VOIP AND MIXEDTRAFFIC.EXPOETIAL BACKOFF ALGORITHM WITH EHACED-RESERVATIO-REQUESTMECHAISMS. 7-10FIGURE 7.9 – BANDWIDTH CONSUMED BY COLLISIONS VS. NO. OF ACTIVE STATIONS FOR VOIP ANDMIXED TRAFFIC. EXPOETIAL BACKOFF ALGORITHM WITHEHACED-RESERVATIO-REQUEST MECHAISMS. 7-11FIGURE 7.10 – BANDWIDTH CONSUMED BY COLLISIONS VS. NO. OF ACTIVE STATIONS FOR VOIPAND MIXED TRAFFIC. SPLITTIG TREE ALGORITHM WITHEHACED-RESERVATIO-REQUEST MECHAISMS. 7-12FIGURE 7.11 – IP HEADER STRUCTURE. 7-14FIGURE 7.12 – MEAN ACCESS DELAY VS. NO. OF ACTIVE STATIONS FOR VOIP AND INTERNETTRAFFIC. SPLITTIG TREE ALGORITHM WITH AND WITHOUT PRIORITISATION. 7-15FIGURE 7.13 – SYSTEM THROUGHPUT VS. NO. OF ACTIVE STATIONS FOR VOIP AND INTERNETTRAFFIC. SPLITTIG TREE ALGORITHM WITH AND WITHOUT PRIORITISATION. 7-16FIGURE 7.14 – CONNECTION-SETUP FOR A FIXED-RATE ACCESS CONNECTION. 7-17FIGURE 7.15 – MEAN ACCESS DELAY AND SYSTEM THROUGHPUT VS. NO. OF ACTIVE STATIONS.SPLITTIG TREE ALGORITHM AND MIXED TRAFFIC. 7-19FIGURE 7.16 – MEAN ACCESS DELAY AND SYSTEM THROUGHPUT. SPLITTIG TREE ALGORITHMAND VOIP TRAFFIC (G.711 AT 64 KBPS). 7-22FIGURE 7.17 – MEAN ACCESS DELAY AND SYSTEM THROUGHPUT. SPLITTIG TREE ALGORITHMAND VOIP TRAFFIC (G.723.1 AT 5.3 KBPS). 7-23FIGURE 7.18 – UPSTREAM VOLUMETRIC DATA AND EFFICIENCY. 7-29FIGURE 7.19 – DVB/DAVIC UPSTREAM PDU STRUCTURES. 7-29FIGURE 7.20 – DOCSIS UPSTREAM PDU STRUCTURES. 7-30FIGURE 7.21– MAXIMUM CHANNEL CAPACITY FOR DVB/DAVIC AND DOCSIS FOR ONE NODE. 7-31FIGURE 7.22 – MEAN ACCESS DELAY AND MAXIMUM CHANNEL CAPACITY, FOR DVB/DAVIC ANDDOCSIS, 12 KBPS STREAMS WITH PACKET SIZE = 1518 BYTES. 7-32FIGURE 7.23 - MEAN ACCESS DELAY AND SYSTEM THROUGHPUT, FOR DVB/DAVIC AND DOCSISWITH 32 KBPS ISOCHRONOUS STREAMS AND DIFFERENT PACKET SIZES. 7-34FIGURE 8.1 – FIXED FIELDS FOR RTP/UDP/IP HEADERS. 8-19FIGURE 8.2 – SIMILARITIES OF CATV WITH FIXED/MOBILE TERRESTRIAL AND SATELLITEWIRELESS ARCHITECTURE. 8-21FIGURE B.1 – PROTOCOL STACK.A-VIFIGURE B.2 – UPSTREAM SLOT AND MINISLOT FRAME STRUCTURE OF THE DVB/DAVIC PROTOCOL. A-VIIFIGURE B.3 – UPSTREAM PDU STRUCTURE OF THE DVB/DAVIC PROTOCOL.A-VIIIFIGURE B.4 – DOWNSTREAM FRAME STRUCTURE.A-IXFIGURE E. 1 – PERFORMANCE OPTIMISATION AND COMPARISON OF THE EXPOETIAL BACKOFFALGORITHM AND THE SPLITTIG TREE ALGORITHM, FOR DIFFERENT TRAFFICPATTERS AND 6 MBPS UP. CHANNEL.A-XIVFIGURE E. 2 – PERFORMANCE ANALYSIS OF ISOCHRONOUS STREAMS FOR DIFFERENT PACKET SIZES, USINGTHE EXPOETIAL BACKOFF ALGORITHM AND A 6 MBPS UPSTREAM CHANNEL. A-XVxiv

LIST OF TABLESTABLE 2.1 – NEXT GENERATION OF CATV SERVICES. 2-9TABLE 2.2 - CERTIFIED CABLE MODEM VENDORS (FROM CABLELABS CERTIFIED/QUALIFIED PRODUCT,INFORMATION AS OF JANUARY 17, 2002 [126] AND CABLE DATACOM NEWS [125]). 2-13TABLE 2.3 – BROADBAND ACCESS TECHNOLOGIES. 2-19TABLE 4.1 – SYMBOLS AND ABBREVIATIONS. 4-18TABLE 4.2 – MAXIMUM THEORETICAL THROUGHPUT FOR DIFFERENT PACKET SIZES. 4-32TABLE 5.1- PACKET SIZE DISTRIBUTION AND CHARACTERISATION. 5-3TABLE 5.2- VOIP ENCAPSULATION WITH AND WITHOUT HEADER SUPPRESSION. 5-5TABLE 5.3 - EFFECTIVE BANDWIDTH IN ISOCHRONOUS STREAMS (KBPS). 5-7TABLE 5.4 – SIMULATION PARAMETERS. 5-8TABLE 5.5 - MAXIMUM SYSTEM PERFORMANCE AND BANDWIDTH CHARACTERISATION. 5-35TABLE 6.1 - MAXIMUM SYSTEM PERFORMANCE FOR INTERNET AND MIXED TRAFFIC SCENARIOS. 6-9TABLE 6.2 – MAXIMUM SYSTEM PERFORMANCE FOR INTERNET TRAFFIC AND DIFFERENT CSAS. 6-12TABLE 7.1 – (OPTIMISED) SIMULATION PARAMETERS. 7-7TABLE 7.2 – VOIP CODECS: G.711 AND G.723.1 7-21TABLE 7.3 – FREQUENCY ALLOCATION AND MODULATION CHARACTERISTICS. 7-26TABLE 7.4 – TECHNICAL CHARACTERISTICS. 7-27TABLE 7.5 – SIMULATION PARAMETERS. 7-28TABLE 7.6 – NUMBER OF STREAMS SUPPORTED FOR DVB/DAVIC AND DOCSIS. 7-35TABLE 8.1 – SUMMARY OF MAXIMUM SYSTEM PERFORMANCE AND BANDWIDTH CHARACTERISATION. 8-7TABLE 8.2- SUMMARY OF MAXIMUM SYSTEM PERFORMANCE FOR DIFFERENT CSAS. 8-9TABLE 8.3 – PRIORITIES FOR CONNECTIONS. 8-19TABLE C. 1 – CABLE TV PENETRATION IN EUROPE, NORTH AMERICA AND ASIA. MAIN SOURCES FROMECCA ANNUAL MEETING 2001 [7] AND EUROPEAN UNION [30].A-Xxv

Chapter 1INTRODUCTION1.1 Statement of the problemThe growing demand for high-speed Internet access and broadband services hasmotivated the extensions of communications services in residential areas [99]. There areseveral technologies currently arriving in homes [124], such as Digital Subscriber Loop(DSL) 1 , Fibre, Wireless and Satellite Technologies. However, Community AntennaTelevision (CATV) is also one of those technologies with high penetration intocommunities (a share of cable TV households subscribers on homes passed by cable)[99]. Belgium, Luxembourg, the Netherlands, Sweden and Switzerland are countrieswhere CATV networks are the most widespread, [7] and [30]. The aggregate number ofcable subscribers in these countries is about 15 million, with a cable penetrationbetween 90 and 100% (the highest percentage in the world), [7]. Germany has moresubscribers with 22 million (at 85% penetration).Historically, cable penetration is weak in Greece, the UK, Portugal, France, Spain andItaly, between 7 and 42% penetration [7]. The reason why the United Kingdom is anexception to this trend is due to satellite penetration. In other parts of the world, theUSA and China have a high cable subscription with 67 million (65% penetration) [128]and 90 million (80% penetration) [135], respectively. Table C.1 (Appendix C) shows anapproximation of the number of cable subscriptions and penetration (in %) for countriesin Europe, North America and Asia.1 A complete list of abbreviations is provided in Appendix A.1-1

Chapter 1Introduction pBecause of high existing penetration combined with high channel capacity, CATVnetworks are at the cutting edge in high-speed communications in homes. For thedelivery of digital information, CATV networks use a spectrum of about 550-870 MHz,that supports about 40 downstream channels in Europe and 50 channels in NorthAmerica. In Europe, the maximum data rate per channel can be up to 50 Mbps, whichresults in an aggregate bandwidth of 2 Gbps. With this capacity, CATV networks cansupport all of the popular communication services (e.g. video and audio on demand,high-speed Internet access, etc) that only require a limited bandwidth in the upstreamchannel for their operation.The available spectrum for the upstream direction is small, ranging from 5-65 MHz inEurope and 5–42 MHz in North America. The maximum number of upstream channelsis reduced to approximately 15 channels (4-MHz channel with a capacity of 6.17 Mbps)in Europe and 11 channels (3.2-MHz channel with a capacity of 10.24 Mbps) in NorthAmerica. This corresponds to a maximum aggregate bandwidth of ≈ 92 and 112 Mbps,respectively.Therefore, the available bandwidth in the upstream direction, makes CATV networkslimited for the support of an increased number of users and particularly when interactiveservices demand a considerable network capacity (e.g. videoconferencing, videotelephony,Voice over IP-VoIP and interactive multimedia).Several technologies help to overcome the problem of limited bandwidth. One is tomake more efficient use of the existing bandwidth by sending signals in the guard bandsthat separate each channel from those next to it. A second method is to compress thesignals. The last strategy to increase the efficient use of the upstream bandwidth is tooptimise or enhance the functionality of the CATV MAC protocol used. In this thesis,we focus on the last technique, which has been the focus of research for the past fewyears.Currently, there are three different MAC protocols for CATV networks. They are theIEEE 802.14, Data Over Cable Service Interface Specification (DOCSIS) and theDigital Video Broadcasting (DVB) / Digital Audio-Visual Council (DAVIC), whichhave been produced by the IEEE 802.14 Working Group, Multimedia Cable Networks1-2

Chapter 1Introduction pSystem (MCNS) Holdings, and the DVB/DAVIC Committee, respectively. Section 2.5presents a detailed description of the standardisation process and the maincharacteristics of each standard.Most of the research found in the literature (described shortly) focuses on performanceevaluation, improvements and optimisations for the IEEE 802.14 and DOCSIS MACprotocols. However, research based on the European cable communication standard(DVB/DAVIC) is scarce. This protocol has a strong support from the European CableCommunication Association (ECCA), EuroCableLabs (ECL) and European cableoperators or Multiple Service Operators (MSOs) as they are more commonly referredto. The major issue for the DVB/DAVIC Committee is the support of IP telephony overcable. This would allow European MSOs to compete with Telephone Companies(Telcos) in the local loop for telephony services and, in general, for high-speed datacommunications services over their existing coaxial infrastructures.The first two versions of the DVB/DAVIC have been finalised (DVB/DAVIC 1.0 [32]and 1.2 [34]). Extensions to the protocol are being added. The major revision came inDVB/DAVIC 1.2, which addressed a splitting tree algorithm (used to resolvecollisions), minislots (used to send shortened reservation requests) and particularlyQuality of Service (QoS) support. More recently, a third version of the DVB/DAVICprotocol was realised in October 2001 [35], which includes new enhanced reservationrequest mechanisms (e.g. piggyback and continuous piggyback) and header suppressionfor Voice over IP (VoIP).Although DOCSIS modems, with European PHY spectrum (referred as EuroDOCSIS[2]) are an alternative for European MSOs, these modems are not preferred for use inEurope [64]. This is because a number of manufacturers are putting investment into theDVB/DAVIC cable modem solutions both for end user and for the interactive networkadaptors in the cable headend.1-3

Chapter 1Introduction pA first step in the provision of data communications services and particularly IPTelephony over the DVB/DAVIC protocol is the implementation of a MAC model, inorder to address capacity planning and quantification of the services that can beprovided to users.The research presented in this dissertation focuses on the DVB/DAVIC MAC protocoland the upstream channel, which combines elements of random access, collisiondetection via the downstream channel, bandwidth allocation for contention access andreservation methods that are not an issue on the downstream direction. This thesisapproaches the performance of the DVB/DAVIC protocol and provides severalenhancements to the MAC layer that improve the efficiency of this protocol for thesupport of timing critical interactive services and high-speed data transmissions.1.2 Modelling of communications networks and researchrelevant to performance evaluation of CATV protocolsIt was not until the evolution of Local Area Networks (with the developments ofEthernet and Token Ring networks) that the performance of communications networksbecame an interesting and important topic [74]. Performance issues arise from a need tosupport multimedia applications on computer networks. This demand has led to theintroduction of techniques and protocols that are designed to handle both constant andvariable bit rate applications simultaneously with high quality [74].1.2.1 Performance considerations of computer networksThe required measurements for the performance analysis of communications networksinclude, among others: Mean access delay; the average delay that occurs between thetime a packet is ready for transmission from a node, until the packet is transmitted andreceived at the final destination. Throughput; the rate of user data being transmittedbetween nodes. The analysis is commonly done in terms of the total number of bitstransferred (kbps, Mbps or as a percentage of the channel capacity) and the letter “S” isnormally used as an abbreviation. Utilisation of the network medium; the fraction ofthe total channel capacity being used, (including data, protocol overheads, reservationrequests, retransmissions and collisions).1-4

Chapter 1Introduction pThese are the most important performance considerations according to [114]. Resultsfor such parameters are generally plotted either as a function of the offered load, whichis the actual load or traffic demand presented in the network, or as a function of thenumber of (active) stations transmitting traffic to the network. In addition, there areseveral other properties that can affect performance. These are:1. Channel capacity 4. Local network protocol2. Propagation delay 5. Offered load3. umber of bits per frame 6. umber of stationsThe first three factors listed above, can be seen as the parameters that characterise thenetwork and are generally treated as constants. The local network protocol is the focalpoint of the design effort and consists mainly of the medium access and physical layers.The physical layer is not likely to be much of a factor. Generally, at this layer datainformation is transmitted with little delay. The medium access layer however does havea significant effect on network performance and is discussed in length in this thesis. Thelast two factors are concerned with determining the performance as a function of offeredload or number of active stations.One factor that was not listed above is the error rate of the channel. With errorcorrection techniques used by communications protocols, such as CATV protocols [52],[22] and [34], link errors are not a significant factor in performance [114]. Therefore,the error rate of the channel will not be mentioned again.1.2.2 Modelling techniquesInitial techniques used to estimate the performance analysis of communicationsprotocols relied on mathematical models using stochastic processes based on probabilityand queuing theory [120]. The high complexity involved in the solutions for queuingnetworks led to the formalisation of approximation methods such as Mean ValueAnalysis (MVA) [89], [90], convolution [91] and linear programming [65] and [66].Such models have made several assumptions, examples of which are random rates ofpacket arrival and fixed number of stations. According to [71], the random arrival1-5

Chapter 1Introduction passumption is incorrect and the traffic in computer communications networks is burstywith distinguishable patterns that are repeated in specific time intervals.Therefore, whilst useful, a random arrival assumption may not be totally accurate in theanalysis of computer communication networks [115].However, even if packet arrivals times could be accurately predicted, networkperformance evaluation is far from solved due to the hybrid access of random accessprotocols.Simulation is another technique used for performance analysis, which has been used toanalyse communications protocols with great success during the last decade [70]. Theuse of simulation must be due in part to the large number of networks in existence.Simulation packages designed specially for communications systems, which reducemodel development and analysis time considerably, have better accuracy and benefitfrom a continuous increase in available processing power. In general, simulation modelsare designed to study more complex scenarios than analytical techniques [4].Errors can be introduced during the design stages and as a check to ameliorate this,simulation results are compared against results obtained from other methods ofperformance evaluation. Methods may include mathematical analysis and/orbenchmarking.A drawback of benchmarking in validation is that this requires an existing system thatcan be measured. Complex network architectures are difficult to benchmark, (e.g. whentrying to predict performance of geographically remote networks).1.2.3 Related work to performance evaluation of CATV protocols1.2.3.1 Analytical modelling of CATV protocolsAnalytical models for the performance evaluation of CATV protocols are few innumber. A difficulty is that CATV protocols are too complex to be modelledanalytically without making an unreasonable number of assumptions. Therefore, it isuncertain whether such analytical methods are useful. CATV protocols, includingDOCSIS, IEEE 802.14 and DVB/DAVIC are random access protocols based on the1-6

Chapter 1Introduction pTDMA technique, which could not consequently be accurately modelled with existingmodels for either TDMA or random access mechanisms.In the area of this dissertation few works with an analytical approach have beenattempted. An exception was the work carried out by Sriram 2 [113]. He proposed asimplified analytical model to estimate the mean packet access delay of the IEEE802.14 and DOCISIS MAC protocols.In order to make a reliable estimation of mean access delay, issues such as contentionaccess delay (D c , the time a station takes to transmit a reservation request successfully)and queuing delays (W q ) should be addressed.For the approximation of queuing delays, according to Gross and Harris [45], thereliable result is given by the method of Pollaczek-Khintchine (PK). The author of[113] used the PK formula for queuing delays and for contention access delays (D c ) 3 , inhis work he used a simplified estimation given by a high percentile value of the meanplus five times the standard deviation. The same method, presented also in Sriram [112],was used to calculate the end-to-end packet delay (D ete ) 4 as the mean access delay plusfive times the standard deviation.We have used a similar approach to estimateDeteof the DVB/DAVIC protocol, and itwas found that the maximum number of users supported is considerably overestimateddue to the high percentile values used. For this reason, it is uncertain whether themethod of [113] can produce reliable results.1.2.3.2 Analysis of the stability of contention resolution algorithmsIn the literature, we have found that studies carried out in the past focused mainly on thestability of contention resolution algorithms (including the exponential backoff,2 The related work presented in this section will be cited the first time by the surname of the authors.3D c= F⋅ ( 2 + 3 ⋅ ρ ) , where F and ρ are the upstream frame duration and channel utilisation.4Dete= Dc+ Wq+ Tx+ 5 ⋅ σ , where TDx is the transmission delay of a packet and σDthe standarddeviation of access delay.1-7

Chapter 1Introduction ppolynomial backoff and the splitting tree algorithm) rather than their performancecharacteristics (e.g. access delay, system throughput and utilisation).a) Exponential backoff algorithm: In studies involving the exponential backoffalgorithm, Aldous [3] has proved that the exponential backoff algorithm is alwaysunstable in the infinite model for any positive arrival rate (λ) 5 . Studies carried out byHåstad et al. [48] also demonstrated that the binary exponential backoff algorithm isunstable for any λ above 0.568 even for a system with a finite numbers of stations.However, a recent study carried out by Goldberg et al. [39] suggested that thestability of the exponential backoff algorithm can possibly be as long as λ < 1/e in theinfinite model and with a finite expected delay (the average waiting time of messagesin the system).b) Polynomial backoff algorithm: Raghavan and Upfal [83] and the authors of [48]proved that this algorithm is always stable for any λ < 1 and for a finite number ofstations. However for the infinite model, Kelly and MacPhee [62] proved that thepolynomial backoff algorithm is always unstable.b) Splitting tree algorithm: Most of the work related to the splitting tree algorithmfocuses on stability and the time required to resolve collisions, rather than estimatingthe total mean access delay. As an example, Greenberg et al. [44] used a hybridalgorithm based on a base 2 estimation algorithm (access probability = 2 -i ) and thesplitting tree algorithm of Capetanakis [14] and Tsybakov [119] to find an estimationof the multiplicity of conflicts ‘n’ and prove that the time to resolve conflicts ofmultiplicity is stable, achieving a maximum throughput = 0.4015. Cidon and Sidi[17] extended the idea of Greenberg and obtained an improved algorithm thatreduces the time to resolve conflicts maintaining stability for all λ up to 0.487.Most of these approaches work under the assumption that the number of nodes → ∞ inorder to support the independence of the transmission probability assumption. However,even if the technique for an infinite number of nodes was to be used, it has been proved5 λ is defined as the arrival rate per unit of time. In this section the unit of time is one slot.1-8

Chapter 1Introduction pthat such solutions could not be applicable for the approximation of networks with afinite number of nodes. Therefore, infinite model results have limited relevance to finitesystems like the ones studies in this text.Particularly, the CATV protocols require approximations of both exponential backoffand splitting tree algorithms, which cannot be accurately modelled using these existinganalytical models due to the variable bandwidth assigned to the contention andreservation access regions from cycle to cycle. Therefore, the high degree ofcomplexity and the need for accurate results indicate the use of simulation techniquesfor the performance evaluation and optimisation of CATV protocols.1.2.3.3 Simulation modelling of CATV protocolsA number of papers reviewed in the literature have focused on simulation techniques toanalyse specific characteristics and performance issues of the IEEE 802.14 and DOCSISprotocols based on HFC networks in general, and as such might not be directlyapplicable to the DVB/DAVIC protocol. The issues addressed include ContentionResolution Algorithms (CRA), Contention Slot Allocators (CSA), scheduling andprioritised mechanisms, performance comparison of CATV protocols, registration afterpower up and performance evaluation and comparison of CATV protocols. Below is areview of the research performed in each field. Special attention is paid to issuesregarding CRAs, CSAs and prioritised mechanisms.a) Contention Resolution Algorithms (CRA): the research carried out in this field dealswith how best to optimise the time to resolve collisions between reservation requesttransmissions. Citta et al. [18], Sala et al. [95], [97] and Golmie et al. [40], [41]analysed and compared the performance of two CRAs (p-persistence and splittingtree algorithm), for the IEEE 802.14 protocol.The p-persistence algorithm resolves collisions by restricting the contending users totransmit in the next contention minislot with probability p. Thus, when a collisionoccurs only a portion of the users involved in the collision transmits in the nextcontention minislot and eventually the collision is resolved.1-9

Chapter 1Introduction pIn general, the exact value of contending users is not known. Sala used the pseudo-Bayesian estimator proposed by Rivest [92] to estimate the number of contendingusers in each contention minislot.The splitting tree algorithm is more complex than that of the p-persistence, due tothe feedback and allocation information transmitted in the downstream channel inevery signalling frame. With this algorithm, all nodes involved in a collision splitinto a number of subsets as introduced in [14]. In series, the first subset transmits,followed by the second and then the remaining subsets. The chances of futurecollisions are reduced by forcing stations that collided in the same slot to retransmitrequests in different slots in the future. A detailed description of this algorithm ispresented in Section 3.4.3.The authors of [41] studied the performance of the splitting tree algorithm using fourdifferent strategies. These were called Free-access, Blocked-access, R-access andT bound -access. In the Free-access strategy the first transmission of requests areallowed to take place on the same minislots used to retransmit collided requests. Newarrival requests are mixed with old or retransmitted requests. In Blocked access newrequests are not allowed in the minislots used to resolve current collisions. This isillustrated by a contention interval that is split into two regions. One region isreserved for ongoing collision resolution and the other, denoted as the newcomerminislot region, is open for newcomer requests.In R-Access, which is also referred to as Adaptive p-persistence in [40], only aportion of newcomer stations are allowed to transmit in the newcomer region. InT bound -access, the headend sets a value (termed as T bound ) every cycle and a station isallowed to transmit in the newcomer region only if the arrival time of the newmessage is less than T bound . Golmie found that by restricting the contention access ofnewcomer arrivals (with R-Access and T bound -access), the performance of the IEEE802.14 could be improved.In was found in [97] that performance difference between the studied algorithms isvery small, with a slight increase in performance in favour to the p-persistencealgorithm over the splitting tree algorithm. This difference was attributed mainly to1-10

Chapter 1Introduction pthe upstream slot structure. In [41], the structure was fixed (the minislots forcontention access were allocated at the beginning of each signalling frame), while theslot structure in [97] was variable (the minislots for contention access were spreadover the signalling frame).b) Contention Slot Allocators (CSA): here the research focuses on the optimisation ofthe number of contention minislots (or request minislots) that should be allocated forcontention access. For the IEEE 802.14 protocol, Sala et al. [96], [98], Lin et al. [73]and the authors of [41] proposed different mechanisms to dynamically allocate thenumber of request minislots for contention access according to the traffic load. Forthe DOCSIS protocol, Cho et al. [15] optimised the number of request minislots bychanging the size of the signalling frame (referred to as MAC Management Access -MAP in DOCSIS), and the number of contention minislots per MAP.In [96] and [98] two mechanisms were studied, Simple and Forced Minislots CSA,which are applied only to the p-persistence CRA. In the simple-CSA mechanism theun-scheduled slots in every signalling frame are allocated as request minislots. In theForced-CSA, the author suggests that in order to obtain optimum systemperformance, the number of contention minislots per reservation request is given bythe maximum throughput of the Slotted Aloha system [116].In [41] two mechanisms were also introduced for contention allocation, Fixed andVariable CSA. In the Fixed-CSA technique the number of request minislots in eachsignalling frame remains the same, whilst in the Variable-CSA technique the numberof contention minislots is dynamically adjusted according to the offered load,maximum request size, number of minislots in a data slot and the frame structure.The results obtained in [41] revealed that the Variable-CSA performs better than theFixed-CSA in terms of lower access delays at higher loads and higher throughput.The authors of [73] have studied several techniques based on the IEEE 802.14 andthe DOCSIS protocol. For the IEEE 802.14 protocol, four different techniques,namely Fix3-Var, Fix3-Fix3, Load-Fix3 and Load-Var were analysed. The first termof each technique (e.g. Fix3 of Fix3-Var) refers to the number of contention1-11

Chapter 1Introduction pminislots to be allocated for new arrivals and the second term (e.g. Var) refers to thenumber of contention minislots to be allocated to resolve collisions.In the Fix3 technique the headend allocates 3 contention minislots in each signallingframe, in the Var technique the headend allocates 3 contention minislots for everycollision registered in the previous signalling frame.In the Load technique, the headend allocated contention minislots, where depends on the current traffic load. From these techniques the author concluded thatthe dynamics of the Load-Var technique allowed the system to perform better thanthe other allocation strategies.For the DOCSIS protocol, the researchers of [73] also proposed different strategies toallocate the optimum number of contention slots in each signalling frame (namely S,E, Dbl, Exp, MeanSE and SE), where S and E represent the minimum and maximumnumber of contention minislots to be allocated, respectively. With Dld the headendallocates ‘max(2·o_Collisions, E)’ contention minislots in the following signallingframe. With Exp the number of contention minislots is given by ‘max(2 o_Collisions , E)’and with SE the headend allocates S contention minislots when there are nocollisions, and otherwise E. From all these strategies the best performance isobtained when the E, MeanSE and SE techniques are used. In the same reference,[73], the authors also approached the number of contention minislots to be allocatedin each signalling frame by changing the backoff window of the exponential backoffalgorithm.The researchers of [15] used a different approach to obtain the optimum number ofcontention minislots. The authors studied the performance of the DOCSIS protocolwhen the MAP size was changed from 1 to 10 ms and the number of contentionminislots was ranged from 2 to 16 minislots per MAP. Optimum results were foundwhen the MAP size was set to 2 ms (which contains 40 minislots of 16 bytes each),and the number of contention minislots per MAP was set to 6.c) Prioritisation and scheduling mechanisms: this field of research focuses on theclassification of data packets (by assigning different levels of priority) to gain accessto the upstream channel at the station premises. At the headend, reservation requests1-12

Chapter 1Introduction pare granted according to the scheduling mechanism adopted. Classification of datapackets (or requests) is needed to support multimedia applications and QoS services,which require relatively low access delays.Limb and Sala [72] studied the performance of a Centralised Priority Reservation(CPR) mechanism for the transmission of multimedia traffic for the IEEE 802.14protocol. In this mechanism a station maintains separate queues for each priority ofdata traffic, and the priority is indicated in the type field of the request message.Requests with higher priority have precedence over requests with lower priority. Theheadend maintains separates queues for each priority and schedules in the first placehigher priority requests before serving a queue of a lower priority.Sdralia et al. [102] also studied the performance of a prioritised-scheduling algorithm(at the headend), using up to 8 levels of priorities for the DOCSIS protocol. The mainimplication of this prioritised-scheduling mechanism is that high-priority traffic willalways take precedence over low-priority traffic. More priority can be given to lowprioritytraffic by scheduling it to have more than one slot into the future. Thistechnique is referred to as Scheduling Advance (SA) in [72] and Ivanovich andZukerman [60].Different techniques can also be used. For example, Nichols and Laubach [78]proposed a scheduler algorithm with three levels of priorities, for the transmission ofConstant Bit Rate (CBR), Committed Information Rate (CIR) and Best Effort traffic(BE). Each of these traffic types has preference over the others with a fixedfrequency (e.g. 20% of channel capacity for CBR, 80% for CIR and 5% for BEtraffic).Sala et al. [94] studied a scheduler algorithm with a Self Clock Fair Queuing (SCFQ)discipline, in which every reservation request is attached with a finish service timeand placed in a queue with increasing order of service time. This finish time isdefined as the deadline time to serve (or grant) the request and is computedaccording to the traffic type characteristics (e.g. station’s data rate, packet size,arrival time) and the round trip delay.1-13

Chapter 1Introduction pThe authors of [73] analysed a scheduler algorithm for the IEEE 802.14 protocol,using three different strategies. In the first strategy (SJF) the shortest jobs (withminimum bandwidth request) are served first. In the second strategy (LJF) the largestjobs take precedence over short jobs, and in the last one (MSJF), the shortest jobs areserved first but the data minislots are allocated into several bursts. From these threestrategies it was found that by adopting SJF lower end-to-end access delays areproduced, but a larger delay for contention access is obtained.Research in this area focuses on the implementation or analysis of schedulingalgorithms at the headend. Corner et al. [20] argue that the headend schedulingalgorithm is not sufficient and that a system at the station’s premises integrated alsowith a scheduler algorithm is needed to efficiently support multimedia traffic andQoS services. The authors proposed a new priority scheme for the IEEE 802.14protocol, which separates and resolves collisions in a priority order.In this new system, a priority usage for each contention minislot is allocated by theheadend. Upon packet arrival, a station initially transmits in minislots exactlymatching their priority level, so the headend knows that all stations participating in aparticular collision are of the same priority level. The headend allocates three slots inthe next signalling frame for each collided request. These slots are reserved forrequests of the same priority as with the first collision.Golmie et al. [43] used the same idea of Corner to evaluate the performance of apriority scheme for the DOCSIS protocol. In this prioritised scheme, Golmieproposed a slight modification to the exponential backoff algorithm (used by theDOCSIS protocol), where the backoff value was set equal to the number ofcontention slots reserved for high priority stations.Both mechanisms ensure that in case of collision, high priority stations retransmitrequests in a timely manner.d) Registration after power-up disruption; In this area of research several strategieshave been proposed to optimise the recovery time after a service disruption (causedby power failures or maintenance services). In Sdralia et al. [103] and [104] anevaluation of the recovery of the DOCSIS 1.0 protocol after a large-scale power1-14

Chapter 1Introduction pfailure is presented, using the default ranging algorithm of the DOCSIS protocol.This algorithm adjusts the transmission power every time a station fails to range withthe headend. The same authors in [106] and [105] proposed a persistent rangingalgorithm, where stations try to retransmit using the same transmission power levelfor a number of attempts. Only after this limit is reached are the power parametersmodified and the process started again. Sdralia showed that the persistence rangingalgorithm performs better than the default ranging algorithm of the DOCSIS 1.0protocol, where a reduction of ≈ 38% was achieved in the recovery time.For the IEEE 802.14 protocol, Sala et al. [93] studied the performance of three CRAs(p-persistence, binary exponential backoff and splitting tree algorithm) in case ofregistration on power-up. The authors found that even in the worst case, the p-persistence algorithm performs better than the splitting tree algorithm up to ≈ 2000stations.e) Performance comparisons of CATV protocols; An investigation of protocol issuesand performance comparisons between IEEE 802.14 vs. DOCSIS can be found inGolmie et al. [42] and Smythe et al. [108]. Performance comparisons betweenDOCSIS vs. DVB/DAVIC are reviewed in Smythe et al. [109], [110], [111] and forcomparisons between “DVB/DAVIC vs. IEEE 802.14 vs. DOCSIS can be found inRangel et al. [84]. Here, the comparisons focus mainly on the capacity of theupstream channel to deliver several traffic types and packet sizes, using differentparameters of configuration.From the literature overview, there is a lack of studies that address the scalability,performance and optimisation of the DVB/DAVIC protocol. The first to report a briefanalysis of the performance of the DVB/DAVIC protocol was Schunke [100]. Thisanalysis focused on the performance of the three access modes of the DVB/DAVICprotocol (contention, reservation and fixed-rate) for the support of CBR (64-kbps with48-byte packets) and bursty traffic (64 kbps with 1.6-Kbytes), using a network size of50 stations and simulation time of 5 seconds. Simulation results revealed that systemperformance directly depends of the access mode that is chosen for the current trafficload. The contention access mode is only suitable for a light traffic load, whereas fixed-1-15

Chapter 1Introduction prate access mode handles latency-sensitive traffic with a constant data rate. The reservedaccess mode only handles the transmission of heavy burst data rate traffic.The performance comparison between DOCSIS vs. DVB/DAVIC introduced in [109],[110] and [111] for the DVB/DAVIC part) is based on the analysis presented in [100].In general, the study carried out by Schunke, whilst valuable, is limited and onlypresents a brief analysis of the DVB/DAVIC protocol (in terms of mean access delays),using a fixed network size of 50 stations and two different traffic types (CBR and burstytraffic). In addition, with such a short simulation period (5 seconds for each simulationrun) the steady state may not be obtained for some configurations, which could lead toinaccurate results.Hence, there is an increased need to estimate the performance of the DVB/DAVICprotocol, under realistic traffic scenarios and configurations that allow reliable analysisof the main characteristics of this protocol.In summary, the lack of studies of the DVB/DAVIC protocol, combined with theincreased demand for high-speed communications services, motivate the analysis of thisprotocol in such detail.1.3 Original contributionsThis research provides a rigorous evaluation of the DVB/DAVIC protocol and presentsfundamental performance characteristics. These characteristics address: the maximumupstream channel capacity; mean access delay; system throughput and utilisationbounds; maximum number of stations (or streams) supported; effects of increasing thestation’s buffer capacity; effects of reducing the signalling frame period; effects ofvarying the bandwidth for contention, reservation and fixed-rate access for datatransmissions; effects of changing the packet size for data transmissions; analysis ofmean access packet delays; and the maximum throughput achievable per station. Allthese performance characteristics are given for realistic traffic loads (including Voiceover IP, Internet traffic, isochronous streams and mixed traffic scenarios), as well asdifferent protocol configurations.1-16

Chapter 1Introduction pThis dissertation also focuses on a rigorous analysis of the two adopted contentionresolutions algorithms of the DVB/DAVIC protocol (exponential backoff algorithm andsplitting tree algorithm). Here, we study in detail performance characteristics,optimisation and implementation issues for each algorithm. Furthermore, threeenhanced contention slot allocators are introduced (Simple-CSA, Forced-CSA andVariable-CSA), which dynamically adjust the bandwidth used for contention access,significantly increasing the system performance when different bounds and traffic loadsare considered.The major contributions of this research are the introduction of novel improvements andnew techniques (presented below), which will enable the DVB/DAVIC MAC protocolto provide the delay requirements optimally for the delivery of delay sensitive servicesand high-speed Internet traffic.In this dissertation, we introduce three novel reservation request techniques, namelyReserved-Request, Unsolicited Grant Slot and Enhanced-Pure Reservation Request thatreduce or avoid the increased risk of collisions during large periods of congestion. Inaddition, we compare the performance of these three mechanisms with the enhancedreservation request strategies introduced in Schunke [101] (Continuous ReservationRequest, Piggyback Request and Continuous Piggyback Request).A second approach for the delivery of delay sensitive services involves Quality ofService (QoS) with a guaranteed delivery. In this research, a prioritised mechanism thatprovides reduced access delays for isochronous streams (during contention access) isdescribed. Additionally, the effect of considering header suppression for the delivery ofVoIP traffic is approached.We also present a performance comparison between the DVB/DAVIC and the DOCSISprotocol, where the major characteristics and the fundamental performance properties ofeach protocol are evaluated. For each analysis we provide saturation points, maximumnumber of streams supported and reasons for inefficiencies.1-17

Chapter 1Introduction p1.4 Overview of this thesisChapter 2 presents relevant theory that includes the evolution and description of CATVnetworks. It describes traditional and modern cabling infrastructures, their differencesand similarities, and presents the changes that have to be made to upgrade CATVnetworks, for the support of bi-directional digital communications. In addition,background material included is the introduction of cable modem technology that showsdifferent cable modem configurations, and identifies the requirements for the nextgeneration of communications services. This chapter also outlines the maincharacteristics of protocols proposed from different standardisation bodies, such asDOCSIS, IEEE 802.14 and DVB/DAVIC, and concludes with an overview ofalternative technologies for high-speed digital access.Chapter 3 describes the main characteristics as well as the architecture, MAC operation,and a description of the exponential backoff algorithm and the splitting tree algorithmadopted by the DVB/DAVIC protocol specification.Chapter 4 presents the structure of OPNET simulation models and describes theCommon Simulation Framework (CSF) model, which is the foundation for an advancedmodel. A comprehensive description of the simulation model for the DVB/DAVICMAC protocol is presented and the theoretical model to be used for analysing theDVB/DAVIC performance and validating the simulation model results is formulated.Chapter 5 looks in detail at the performance of the DVB/DAVIC protocol using thesimulation model presented in chapter four. A number of protocol configurations areevaluated for different traffic situations. The main issues addressed in this chapter arethe network capacity, scalability in terms global offered load and number of stations,buffer capacity, signalling frame period and packet size variation, as well as the effectsof changing the bandwidth for contention and reservation access. Results focus onsystem throughput, mean packet access delays and system utilisation when the loadgenerated by all stations is increased up to maximum network capacity.1-18

Chapter 1Introduction pChapter 6 covers a detailed analysis of the dynamics of the two CRAs adopted by theDVB/DAVIC protocol. The analysis focuses on performance optimisations whendifferent backoff bounds for the exponential backoff algorithm and different values forthe Entry-Spreading factor of the splitting tree algorithm are considered. In addition, aperformance comparison between these two algorithms is presented. This chapter alsointroduces three enhanced CSAs (Simple-CSA, Forced-CSA and Variable-CSA) that willfurther optimise the performance of the DVB/DAVIC for different traffic configurations(e.g. Internet traffic, VoIP traffic and mixed traffic).In Chapter 7, three novel reservation request mechanisms are proposed (Reserved-Request, Unsolicited Grant Slot and Enhanced-Pure Reservation Request) for thesupport of timing critical interactive services, and their performance is compared withthe default reservation request mechanism and the three techniques presented in [101].This chapter also describes a prioritisation mechanism that will further reduce the delaycaused by contention access. In addition, the effects of using fixed-rate access andheader suppression for delay sensitive applications are studied, especially for thedelivery of VoIP streams. Finally a performance comparison between DVB/DAVIC andDOCSIS protocol is addressed.Chapter 8 presents a comprehensive analysis of the key findings, and shows how theresults could be used by either vendors or operators in order to improve networkperformance. Finally, issues of further research are outlined.1-19

Chapter 2OVERVIEW OF CURRENT CATVNETWORKS2.1 IntroductionThe main function of traditional Community Antenna Television (CATV) networks –distribution of TV and radio programs- has rapidly been extended in the last few years.Modern cable systems are being designed to deliver a variety of communicationsservices with support for current and future applicationsAn increased growth of the Internet has created a demand for broadband access.Telephone companies (Telcos) have until recently provided data communicationschannels. They were slow and relatively expensive.CATV networks are considered as an alternative bearer of new interactive services,because they have large unused bandwidth. Advances in CATV network technology andin particular the introduction of cable modem technology over Hybrid Fibre Coax(HFC), allows for diverse range of data communications services, such as interactivetelevision, broadband Internet and IP telephony.This chapter presents an overview of CATV networks. The physical characteristics oftraditional cable networks are shown and the structure of modern CATV networkshighlighted. It also presents some of the most important protocol specifications. Specialattention is paid to cable modem technology. Finally, an outline of high-speed DSL andfibre technologies is presented.2-1

Chapter 2Overview of current CATV network p2.2 Evolution of CATV networks and cable modemtechnologyCommunity Antenna Television (CATV) began in 1948 in Astoria, Oregon. It wascreated as a way to improve television reception for people who lived in remote or hillyareas, where good television signal reception was difficult. Since that time, cabletelevision has become a common source of video entertainment for more than 500million viewers around the world [125]. Early interest in cable modem technologyemerged in the 1980s and was focused around the IEEE 802.4 token bus over CATVnetworks.At the same time, research institutions and universities began using the CATVinfrastructure for two-way campus data networking. These institutional networks werecalled I-Nets and were mainly used to connect institutional Local Area Networks(LAN). Unfortunately, such cable modem products were relatively costly tomanufacture and maintain and the resulting broadband data networks were expensive tooperate [63]. Technical obstacles combined with a lack of financial incentive restricteddevelopment. In essence, large-scale investment was required to serve sufficientcustomers for good return.The first organisation to partly overcome these technical difficulties was LANcityCorporation of Andover, Mass. USA in 1990, which became a commercial success withthe development of cable modem technology. LANcity’s cable modems became anessential device as it allowed the CATV networks to extend communications services toentire cities. Soon after LANcity developed its cable modem, Zenith introduced asimilar type of cable modem. By 1992, both of these companies were offering“Symmetrical LAN over cable” [63]. Since then, diverse groups have worked togetherto help cable modem technology and cable data networks standards. Section 2.5presents a number of the CATV standards that support this technology.2-2

Chapter 2Overview of current CATV network p2.3 CATV networksTraditional CATV networks are based on a tree and branch network architecture asillustrated in Figure 2.1. The main components are: headend, trunk cables, amplifiers,feeder or distribution cables, splitters, drop cables, taps and terminal equipment.Signals emanate from a headend location, which receives programming on TV channelsfrom a variety of sources such as satellites, broadcast transmission and local televisionstudios. From the headend the signals are delivered to subscribers, first down a trunk(tree) cable, that carry the signals to residential areas. This trunk cable was historically ahigh quality coax cable and was intended to cover large distances, often well over 10miles and with ≈ 20-40 amplifiers in cascade from the headend to subscriber [24].Along the coaxial trunk route, signals would need to be amplified to retain quality.Amplifiers are required approximately every 0.4 miles depending on the bandwidth ofthe system. The more splits there are in a cable network and the greater the distancefrom the headend, the more amplifiers are needed.The distribution or feeder cable expands around the neighbourhood. The length of thiscable can be up to 1 mile. This limitation is due to the fact that Radio Frequency (RF)energy is tapped off to feed homes.CATV Networkhome dropBroadcast TVTrunk< 16 milesone-wayFeeder/Distribution Cable< 1 miletapdistributioncoax cableHeadendcoax trunkamplifieramplifiersplitterresidential homeSet-Top-BoxTVFigure 2.1 – Example block diagram of a typical CATV network.2-3

Chapter 2Overview of current CATV network pSplitters are used at every junction to distribute the signal between branches. Theinterface between truck cables and feeders is done by bridge amplifiers. Moresophisticated amplifiers are used in the distribution loop because noise levels are higherin the feeder cables than in the truck portion. These amplifiers are called line extendersand operate at higher power levels in this part of the system. Line extenders are requiredapproximately every 0.25 miles and are restricted to a maximum of three [99].Taps to connect subscribers are placed approximately every 150 feet. The drop cableconnects the subscriber to the tap and its maximum length is approximately 400 feet.Terminal equipment is the last component of the system. They are the receivers that usethe signal.As the service delivered to subscribers was originally designed exclusively toaccommodate television programming, the service was unidirectional. The signalsdelivered were analogue and replicas of the one a broadcaster sends through theairwaves. Because of this there was no need to modify the television set.2.3.1 CATV spectrum allocationThe spectrum for CATV networks is divided into the downstream and upstreamfrequencies. Figure 2.2 shows the basic downstream and upstream spectrum allocationsand the slight variations in the frequency ranges for Europe (EU), United States (US)and Japan (JP) [109].Upstream digitalchannelsDownstream analoguechannelsDownstream digitalchannels1 2 ...ncrossover region1 2 3 ... m542 (US)55 (JP)65 (EU)88 (US)90 (JP)110 (EU)1 – 6 MHz (plus 200 kHz in EU)1.6 – 6 Mbps (plus 10 Mbps in US)Figure 2.2 – Spectrum allocation on a CATV network.450869 (US)770 (JP)862 (EU)MHz1 – 6 MHz (US, JP)1 – 8 MHz (EU)28 – 40 Mbps (plus 50 Mbps in EU)2-4

Chapter 2Overview of current CATV network pThe downstream channels support the legacy analogue broadcast television (80-450MHz, the frequency depends on the country), and multiples of 1-6 MHz or 1-8 MHzchannels in the 450-900 MHz region for the transmission of digital data. The upstreamchannels are also divided into 1-6 MHz channels, but the high level of noise means thatdata capacity is only 1-6 Mbps per channel in EU and 1-10 Mbps in US and Japan asopposed to the 28-40 Mbps available in each downstream channel in US and Japan. Adata rate of up to 50 Mbps can also be supported in EU.In terms of modulation schemes, most manufacturers have implemented 64 and 256Quadrature Amplitude Modulation (QAM) for the downstream and Quaternary PhaseShift Keying (QPSK) for the upstream channels. Japan and US have also implemented a16-QAM modulation. These modulation techniques are described in [6].2.3.2 Upgrade of CATV networks to bi-directional HFC architecturesIn the last few years, CATV networks have been upgraded to improve the quality of thesignal transmission and to increase available capacity. Modern CATV networks areinitially built with Hybrid Fibre Coax technology [64]. The gradual evolution has shownthe technical and economical viability of two-way communication in a CATV network.A combined fibre coaxial CATV network is referred to as an HFC System. Fibre linksare used to transport subcarrier multiplexed signals typical to a group of between 500and 2000 subscribers. Such networks are now standard and provide a bandwidth up to750 MHz (in most systems) in the downstream direction for digital and analoguetransmissions [27]. This bandwidth will be up to 1 GHz in the future [12].In order to deliver data over HFC networks, laser transmitters convert a fixed frequency(6 MHz wide in the US and 8 MHz in Europe) sent from the headend into opticalsignals. At the outskirts of a community, a laser receiver, named as Optical NetworkUnit (ONU), reconverts the signals so that they can again be transmitted over thecoaxial cable, which goes into each individual house. At the Customer PremiseEquipment (CPE) a receiver (set-top box or cable modem) tunes to the appropriate fixedfrequency in order to receive downstream signals.2-5

Chapter 2Overview of current CATV network pThe support of digital signals requires the use of the appropriate modulation technique.The most widely used digital modulations for HFC networks are QAM, QPSK andVestigial Side Band (VSB).A modern CATV network for analogue video and digital services is presented inFigure 2.3. The truck cables from traditional CATV networks are replaced with highreliable,low-attenuation fibre links. This implies that a large number of the analogueamplifiers along the tree trunk are no longer needed. Hence, the reliability of the systemincreases and the quality of the signal improves, since there exist fewer activecomponents between the headend and the subscribers [12].In addition, one-way (line extender) amplifiers from traditional CATV networks needalso to be replaced by two-way amplifiers to allow upstream transmission fromsubscribers to the headend. Controllers at the headend, (referred as Cable ModemTerminations Systems -CMTS or Interactive Network Adaptor –INA) and cablemodemsneed to be installed at the headend and customer premises, respectively.Broadcast TVModern CATV NetworkFibre (Trunk)Domain< 100 milesFeeder/Distribution Cable< 1 miletaphome dropPSTNHeadendfibre trunkONUdistributioncoax cableISPInternetCATVONUtwo-wayRouterATMswitchCMTS/IAamplifiersplitterCustumer Premise EquipmetLAServerFile ServerVideo ServerresidentialhomeSet-TopBoxCable ModemFigure 2.3 – Modern CATV network with HFC architecture.2-6

Chapter 2Overview of current CATV network p2.4 Cable modem technologyDigital data signals are carried over Radio Frequency (RF) carrier signals on a cablesystem. Digital data utilises cable modems, devices that convert digital information intoa modulated RF signal and convert RF signals back to digital information. Theconversion is performed at the subscriber’s premises, and again by the headendequipment handling multiple subscribers.Cable modems have some advantages over other competing technologies, such as dialupmodem and the digital subscriber line (DSL). Although cable modems can operate atspeeds up to 50 Mbps downstream and 12 Mbps upstream [22] and [35], they arenormally programmed to operate at about 500 kbps - 2 Mbps upstream and ≈ 500 kbps -10 Mbps downstream. Subscribers are always connected, eliminating the call set-uptimes. Some of the disadvantages are that cable modems are less secure than DSLmodems because the line is shared with others in the neighbourhood and data speedsvary according to the number of active users.2.4.1 Cable modem configurationsCurrently there are three configurations for cable modems: external cable modems,internal cable modems, and interactive set-top boxes (STB), as illustrated in Figure 2.4.• External cable modem configurations are common and there are severalcombinations. Four examples are given here. The first combination is to have a PC,Cable Modem Termination System/Interactive Network AdaptorCableModemCable ModemRouter/Bridge/SwitchNIC CardNIC CardNIC Card(Cable Modem)Set-top-box10/100Base HubInteractive -TVExternal Cable ModemInternal Cable ModemSet-top-boxFigure 2.4 – Cable modem configurations.2-7

Chapter 2Overview of current CATV network pwith a cheap Network Interface Card (NIC) inside, directly connected to a cablemodem through an Ethernet connection (10BaseT) or Universal Serial Bus (USB).The second alternative is to have two PCs connected to a single cable modem. Mostproviders have instructions on how to accomplish this and require a home user todownload additional software (such as Network Address Translator-NAT [107]) toaccommodate the dual connection on a single modem [9]. This means that the MACaddress of the packets coming from a PC (without cable modem connection) shouldbe changed to the MAC address of the PC’s NIC card with cable modemconnectivity. This is because most cable modems only forward traffic to the CATVnetwork from a single MAC address.The third alternative is to have a router, bridge or switch connected to the cablemodem. With this configuration, the cable modem is attached to the network and therouter/bridge/switch is responsible for forwarding traffic onto the cable system usingits MAC address. This is an attractive solution but it is more expensive because acostly device is required.A last configuration presented is to have two NIC cards in a single PC and thecorresponding software (NAT) to forward traffic from other PCs. With one NIC cardthis PC has connectivity with the cable modem. With the other NIC card, a 10/100Base Hub can be connected, which in turn interacts with others PCs. This is aninexpensive solution and allows multiple PCs to have interconnectivity with theoutside world.According to [9], the use of the PC with the two NIC cards is not necessary since thehub can be connected directly to a cable modem. The hub will act as a bridgingfunction onto the cable modem, and concentrate the traffic through the individualdevices.• Internal cable modems are cable modems that are built into a PCI add-on card. Itmay be the cheapest implementation possible, but there are some drawbacks. Thedesign only allows a cable modem to connect to one computer. In addition, internalcable modems do not isolate the connection to the cable network from the AC powersystem. This can cause major problems in some cable systems.2-8

Chapter 2Overview of current CATV network p• The Interactive STB configuration serves a very different function to the first twoconfigurations. The STB is not really designed to connect to a computer, although itfunctions in a similar way to a cable modem. Its primary role is to provide moretelevision channels on the same limited number of frequencies. This is made possibleby using Digital Video Broadcasting (DVB) technology. STBs also provide two-waycommunication. This two-way communication allows the user access to Enhanced-TV, Interactive TV, Web-browsing, email, etc., directly on the TV screen.2.4.2 Next generation servicesThe services that modern CATV networks would provide are listed in Table 2.1(Sources from [16], [36] and [25]). Not all of these services have to be supported in thenear future, but the network infrastructure should be able to support all of them as andwhen demand rises. Most of these new services are graphics, audio and video oriented,and contrary to the simple text-oriented first generation application, they demand ahigh-speed communications infrastructure.In addition, the requirement of ‘real-time’ interactivity makes the need for multimegabit-per-secondhigh-speed communications even greater. Since voice and videotransmission are inherently time sensitive (isochronous) applications, the network has toexhibit deterministic behaviour, and there has to be an upper limit for the transmissionAnalogue and DigitalTV and AudioTable 2.1 – Next generation of CATV services.TelecommunicationsServicesMultimedia ServicesLocal ServicesPay-per-channel Telephony Online-services Local-channelsPay-per-view Video Phone Services Interactive games Local TVAudio-on-demand Videoconferencing Distance Learning Local RadioVideo-on-demand Fax Interactive Video Local online servicesHigh-definition TV Telemetry Interactive MultimediaInformation ServicesNear-video-on-demandTele-banking/shoppingTele-teaching learningShopping and bill payingservicesHigh-speed Internet accessLAN interconnection(Leased Lines D1/E1)GamblingRemote-control andmonitoring2-9

Chapter 2Overview of current CATV network pdelay of chunks of audio and video. Bounded transmission delay is vital, since itdictates buffer size within the receiver and transmitter as well as the quality of thetransmission.For instance, for a videoconferencing service, the maximum tolerable (one-way) meanaccess delay for 95% of all video frames transmitted is ≈ 300 ms in order to guarantee agood quality of human interaction [38]. However, voice streams require lower accessdelays. As an example, high quality VoIP frames using a G.711 codec require a delay ofbetween 1-20 ms and controllers for acoustic and electric echo are not needed, [118].This delay can be more flexible for VoIP using codec G.723.1, which can be up to 150ms for a good quality call. Here echo control is required but does not compromise theeffective interaction between the users. If the delays are in the range 200 to 400 ms, theeffectiveness of the interaction is lower but can be still acceptable, [118].2.5 Standards for CATV networksThe development of standards and common protocols is essential to any wide-spreadacceptance of similar technology. Standards are needed for new services so thatdifferent vendors' products can interoperate. Standards will provide economies of scalefor a mass market resulting in lower cost products, because many internal elementswould have the same specifications and could be manufactured in huge quantitiesTo make this vision true, industry groups, including the Digital Audio Visual Council(DAVIC) [129], Digital Video Broadcasting (DVB) [130], EuropeanTelecommunications Standards Institute (ETSI) [132], Institute of Electrical andElectronics Engineering (IEEE 802.14), ATM Forum [5], Multimedia Cable NetworkSystems (MCNS) Holdings [127], and the Society of Cable TelecommunicationsEngineers (SCTE) [134], created specification with the goal of increasinginteroperability between equipment.Alliances between the different groups led one organisation to adopt the other standards.As an example, the SCTE adopted the MCNS specification and the DVB group adoptedDAVIC 1.2, while the IEEE 802.14 decided to work with the ATM Forum for theimplementation of ATM over HFC networks.2-10

Chapter 2Overview of current CATV network pCommon goals within the above groups have resulted in the development of severalinternational standards for CATV networks. A comprehensive description of thesestandards can be found in [84] and [109]. We list the main protocols for cable systemshere.• DVB/DAVIC: Digital Video Broadcasting (DVB) which has adopted the DigitalAudio-Visual Council (DAVIC) recommendations with respect to CATV, has beenresponsible for the development of the European standard ETSI ES 200 800 [34],widely known as the DVB/DAVIC protocol specification (DVB-RCC). Chapter 3presents a detailed description of this protocol.• DOCSIS: The Data Over Cable Service Interface Specification (DOCSIS) standardemerged from the work of MCNS, a consortium consisting of Comcast, TimeWarner, TCI, and Cox Communications. MCNS in turn partnered with MediaOne(previously Continental Cablevision), Rogers Cablesystems (Canada) and withCableLabs [127] who have now administered the initiative since 1997. Thispartnering of interests represents operators that provide cable service to a majority ofNorth American cable subscribers, with connected homes in excess of 60 million.(85% in the US, 80% in Canada and 12% in Mexico), [8].CableLabs has produced the Data Over Cable Service Interface Specifications,which comprise DOCSIS 1.0 [21], DOCSIS 1.1 [22], DOCSIS 1.2 [23] andEuroDOCSIS [2], on behalf of the North American Cable industry and using cablemodem technology.DOCSIS 1.0 is an evolved LANcity-based protocol, targeted at residential, low cost,off-the-shelf cable modems with certified interoperability between vendors, [69].The initial release was in December 1996. The basic architecture of this protocol is asingle, large Ethernet-based bridged LAN with primarily a best-effort Internet accesssystem. It was not designed for QoS support.DOCSIS 1.1 is based upon the required needs of the PacketCable Audio/VideoCodecs Specification [80]. The PacketCable project [133] is an internal projectwithin CableLabs and its main task is the support of voice and video applications.2-11

Chapter 2Overview of current CATV network pThe DOCSIS 1.1 version added substantial protocol support to provide dynamicQoS features for packet voice services, in addition to packet data services. Otherenhancements include, for example, baseline privacy and multicast support.DOCSIS 1.2 is a detailed technical specification of a next-generation physical layertechnology (PHY) for integration into the Data-Over-Cable Service InterfaceSpecification. The protocol specification is based on Terayon's Synchronous CodeDivision Multiple Access (S-CDMA) and Broadcom's advanced Frequency AgileTDMA (FA-TDMA), which were first integrated into the IEEE 802.14 protocolspecification [54]. These modulation techniques will allow cable modems to supportan upstream data rate up to 30 Mbps. This version recently became officiallyaccepted as the DOCSIS 1.2 protocol specification [23], in December 2001.Today, CableLabs runs a DOCSIS vendor certification process for cable modems. InJanuary 2002, more than 100 cable modems were DOCSIS 1.0 certified fromapproximately 80 manufacturers.• IEEE 802.14: The IEEE 802 was responsible for the development of LAN/MANprotocol standards. In May 1994, the IEEE 802.14 sub-committee was established todevelop a MAC and PHY specification to support cable networks. This effortresulted in the development of the IEEE 802.14 protocol specification. In July 1997the committee released the first internal draft for the MAC and PHY layers [51].However, MCNS' effort undermined the IEEE 802.14 group's work and was able todefine a specification much quicker than the IEEE. Despite the launch of the NorthAmerican initiative, the IEEE 802.14 committee continued its work with theobjective of creating an international, rather than national, standard.Unfortunately, in September 1999, the joint effort was ceased, followed by thedisbanding of the IEEE 802.14 Working Group in November 1999. This was theresult of the slow standardisation process within the IEEE, which failed to observethe time to market constraint and lost the support of industry [84]. The latest draftspecification, Draft 3, Revision 3 [52], will remain as a proposed specificationwithin IEEE for one year after which time, if there is no interest by any other groupor body, not necessarily within IEEE, it will be withdrawn.2-12

Chapter 2Overview of current CATV network pIn an effort to acknowledge the IEEE 802.14 Working Group's work, CableLabshave implemented IEEE 802.14's advanced PHY specification [54] within DOCSIS1.2 [23]- a partial victory for the group.2.6 Certified cable modemsThe first organizations that offered proprietary cable modems and helped to release thecable modem’s potential were LANcity, Zenith and Intel in 1993. Two years later,Motorola, Hybrid, Com21 and Westend joined in the development of this technology[125]. Today, approximately 80 vendors have entered the cable modem revolution,which is now a multi-billion-dollar industry. Of these vendors only ≈ 70 have obtainedat least one DOCSIS CM or CMTS product certified by CableLabs [127].Table 2.2 shows the current cable modem manufacturers with certification fromCableLabs. This table also presents the organisations manufacturing DVB/DAVICequipment.Table 2.2 - Certified cable modem vendors (from CableLabs Certified/QualifiedProduct, information as of January 17, 2002 [127] and Cable Datacom News [126]).DOCSIS-ProductsDOCSIS 1.0: CM (0)DOCSIS 1.1: CM (1)DOCSIS 1.0: CMTS (2)DOCSIS 1.1: CMTS (3)EuroDOCSIS 1.0: CM (4)EuroDOCSIS 1.0:CMTS(5)DVB/DAVIC-ProductsDVB/DAVIC: 1.0 EM (0)DVB/DAVIC: 1.0 INA (1)DOCSIS-Vendors with CableLabs Certification3COM(0,2,45), Aastra(0), Accton(0), ADC(2,5), Alcatel(0),Ambit Microsystem(0,1), Arris(0,1,2,3,4,5), Askey(0,4), Asustec(0,4),BAS(2), BestData(0), CIS Tecnology(0), Cadant(3), CastleNet(0,4),Cisco Systems(0,2,3,5), Com21(0), Coresma(0), Correlant(0),D-Link Systems(0,4), DX Antenna(0), Dakos(0), Dassault(0,4),DeltaKbel(0), ElsaAG(0,4), Ericsson(0,1,4), Future Networks(0),General Instrument (0), GVC(0), Global Telemann(0), Hauppauge(0),HighSpeed Surfing(0), Hitron Technologies (0), Infinite(0),Joohong(0), LG Innotec(0), LinkSys(0), Lucent(Delta-Kbel)(0), MasproDenkoh (0), Matsushita Panasonic(0), Motorola(0,2,4,5), NetGame(0),NetGear(0), Net & Sys Co.(0,4), Nortel(0), Ole Communications(0),Pacific BB Comms(2), Panasonic(0), Philips(0), Powercom (0),RiverDelta(2), RiverStone(2,5), SMC Networks(0), SOHOware(0),Saejin (0), Samsung (0,4), Scientific-Atlanta(0,1,4), Sony(0), TCE(0),Techno Trend (4), Tellabs(0,1,2,5), Terayon(0,2,5), TexasInstrumets(0,1), Thomson(0,4), Toshiba (0,1,4), TriGem(0),TurboComm(0), U.S. Robotics(0), Zoom Telephonics (0), ZyXEL (0).DVB/DAVIC -VendorsAlcatel (1), Cisco Systems(1+DDIC), Com21 (0+DDIC), DeltaKbel(0),HB Telecom(0), Hughes Network Systems (0), RiverDelta(1),Terayon(0+DDIC), The Industree (0+DDIC), Thomson (1+DDIC)CATV vendors in bold have obtained more than two cable modem devices certified by CableLabs.2-13

Chapter 2Overview of current CATV network pHowever, not all DVB/DAVIC vendors have obtained the verification from theDVB/DAVIC Interoperability Consortium (DDIC). Moreover, the number oforganisations producing DOCSIS equipment is considerably large when compared withthe number of DVB/DAVIC vendors (≈ 10).Most cable modems, currently deployed as a commercial service, are external cablemodems (with Ethernet or/and USB ports). Some cable modems have advancedcapabilities. For instance, Motorola and Com21 have manufactured external cablemodem routers that have a 4-port hub, built-in NAT, Dynamic Host ConfigurationProtocol (DHCP) server and firewall for creating home networks. Interactive, Toshiba,Tellabs and Cisco Systems have produced external cable modems with PacketCableTechnology [80] that allows VoIP calls (and may be capable of video over IPtransmissions in the future).Moreover, Terayon is the first organisation that has already produced cable modemscompliant to DOCCSIS 1.2 [23], which includes advanced PHY capabilities and dualmodulation techniques, S-CDMA and FA-TDMA. DOCSIS 1.2 triples the upstreamcapacity of DOCSIS 1.1 (up to 30 Mbps with 64-QAM modulation) enabling operatorsto create new services for residential and business markets, such as video conferencingand peer-to-peer applications.Currently, the number of cable modems shipped worldwide has reached the top-mark of15 million modems, with Motorola, Toshiba and Thomson leading the market with39%, 17% and 12% of the sales, respectively [126].2.7 Competing technologiesMultiple broadband competing technologies are being developed to provide high-speedaccess to small office and residential areas. Digital Subscriber Line (xDSL), and severalforms of passive optical networks (FTTx) are direct competitors to HFC infrastructuresand the challenge is to work out the least expensive and most efficient solution, whiletaking into account the different boundary conditions. In this section, a review of thesecompeting technologies is addressed.2-14

Chapter 2Overview of current CATV network p2.7.1 FTTx technologiesFTTx is a series of networking technologies that connect the users and the central officevia optical fibre links. Although the cost of optical fibre is not much different fromcoaxial cable, optical equipment is very expensive. There are two main forms of FTTxtechnology:• FTTC: Fibre-to-the-Curb (FTTC) is also referred as Switched Digital Video (SDV).This technology makes use of fibre optic to connect the headend to Optical NetworkUnits (ONU) ‘at the curb’ and serves small groups of homes (between 8 and 24homes) [67]. Twisted pair and coax cable is used to connect the ONU to individualhomes. Twisted pair supports POTS while coax supports analogue video and newdigital services can be supported by either.While TDM is used at present to multiplex signals for delivery to individual homes,ATM switches will most likely replace the ONU in the near future [67]. Telephony,video and digital services are multiplexed for transport over the FTTC network andare de-multiplexed by the ONUs. The maximum downstream bandwidth supportedby FTTC networks is 51 Mbps using coax cable. Twisted pair can also be used tosupport the same bandwidth if the length is not greater than 500 feet.• FTTH: Fibre-to-the-Home (FTTH). This technology is a point-to-point networkarchitecture that uses only fibre links. The cost of deployment of a FTTH network isconsidered high compared to FTTC and HFC. Many people believe that FTTCtechnology is more cost-effective than FTTH due to maintenance and the largenumber of ONUs. However the opposite is true, [76]. Although FTTH networksavoid the use of ONUs, the high number of laser transceivers used and the cost of thefibre links, which run down to the last mile into each home, makes them a costlysolution. However, the increased bandwidth available to each fibre out of theHeadend might be as high as 155 Mbps (OC-3c), allowing each home to become aservice provider by attaching servers within the subscriber’s premises [67]. FTTH isthe topology that offers the highest capacity due to its all-fibre structure and is anattractive solution for Telco’s. Some Telco’s have already installed FTTH networks,such as Nippon Telegraph and Telephone, (NTT) [122] and Deutsche Telecom [123].2-15

Chapter 2Overview of current CATV network pDue to a high installation cost involved, the initial structure may not be point-topoint.However, using a Passive Optical Network (PON) and passive optical couplersto split fibres from the Headend to the home into multiple fibres reduces the numberof laser transceivers required for every home, and in turn reduces the cost ofdeployment. Bi-directional communication over FTTH networks is achieved by aMAC protocol that shares bandwidth on the PON structure among different homeservers [75].• SuperPONs: Super Passive Optical Networks (SuperPONs) is another alternativetechnology in its early stages, which can be applied to FTTC and FTTH networks[81]. This technology is considered as the next generation PONs, designed to rungreater distances (about 60 miles) and can support a larger number of users (up to2048 ONUs). SuperPONs provide shared bandwidth up to 2.5G bps downstreamusing TDM and 311 Mbps upstream with TDMA. The major technical challenge insuch architectures is the development of a MAC protocol that could allow fair accessto the network by such a large number of users.2.7.2 xDSL technologiesxDSL is a series of Digital Subscriber Line technologies which allows for thetransmission of information (voice, video and data), over existing copper telephonelines at high speeds. The xDSL technology is attractive, since there exists a hugeinstalled base of twisted pair lines (800-million) worldwide [26]. Optical fibre will carrysignals from a central office to a neighbourhood node, which in turn converts the signalsand puts them into a telephone line to the user. In some cases, a direct copper line canbe used if the distance from the central office to the user is not too large (usually under4 miles). DSL has a similar problem to cable modem technology; upstream bandwidth ismuch smaller than downstream (except its symmetric system which is much moreexpensive). There are various forms of DSL, referred to as xDSL.• ADSL: Asymmetric Digital Subscriber Line (ADSL) is quickly becoming the mostpopular form of xDSL. ADSL can support up to 8 Mbps bandwidth for downloadingand up to 640 kbps for uploading [6].2-16

Chapter 2Overview of current CATV network pThe asymmetrical nature of ADSL technology makes it ideal for Internet/Intranetsurfing, video-on-demand and remote LAN access.ADSL requires a voice/data splitter, commonly called a POTS Splitter (Plain OldTelephone Service) to be installed at the consumer's home or business location. Thesplitter separates voice from data transmissions. For simultaneous use of thetelephone and data access, additional phone wires may need to be installed at thelocation. Full rate ADSL provides service up to a maximum range of 12,000 feet(about 2 miles) from the provider company's central office to the end-user. Fordistances up to 15,000–18,000 feet (about 2.8-3.4 miles) the data rate decreases to1.5 Mbps downstream and 64 kbps upstream [68].• G.Lite: This technology, often called ADSL Lite, Splitterless or Universal ADSLand now also known as G.992.2 [58], does not require a POTS splitter to be installedat the consumer's home or business. ADSL Lite provides bandwidth downstream upto 1.5 Mbps and upstream up to 512 kbps. ADSL Lite provides service up to amaximum range of 12,000 feet (about 2.0 miles) from the central office. Under goodhome conditions and loop quality the range can be extended up 18,000 feet (≈ 3.4miles) [68]. This technology will be primarily targeted towards residential customers,for combined data and circuit/IP voice services over a single twisted copper looppair. In addition, it is expected that the data rates of this standard will meet the needsof the average consumer for some time to come [9].• SDSL: Symmetrical Digital Subscriber Line (SDSL) delivers high-speed datanetworking over a single-pair of copper phone lines, at the same speed in both theupstream and downstream directions. Speed ranges achieved by this technology are384 kbps, 768 kbps, 1 Mbps, 1.544 Mbps (T1 service over two copper pairs) or up to2.048 Mbps (for E1 streams over three copper pairs) at a maximum range of 12,000feet (about 2.3 miles) [11]. SDSL is ideal for business applications that requireidentical downstream and upstream speeds, such as video conferencing orcollaborative computing as well as similar applications appropriate for ADSLtechnology. SDSL uses either CAP modulation or the same kind of line-modulationtechnique employed in ISDN, known as 2B1Q.2-17

Chapter 2Overview of current CATV network p• HDSL: High-data-rata DSL (HDSL) is a two or three copper pair technology thatachieves symmetrical data rate transmissions, conforming to T1 (1.544 Mbps) or E1(2.048 Mbps) standards, respectively [37]. This technology uses either baseband2B1Q or passband CAP modulation schemes and the distance from the central officeis limited to 12,000 feet (≈ 2.3 miles). According to [9], the use of two or threecopper pairs is no longer needed, because the most recent versions of HDSLarchitectures use only one pair of wires, and it is expected to be more accepted by theproviders.• VDSL: Very high bit-rate Digital Subscriber Line (VDSL) is the fastest xDSLtechnology. Several VDSL formats have been proposed and trailed. However,standardisation is still in process at ETSI. This technology emerged in order todeliver multi-megabit data rates over short spans of copper wire, such as thedistribution of digital TV programming to the neighbourhood node for FTTCapplications.VDSL delivers up to 52 Mbps downstream and from 1.5 to 2.3 Mbps upstream overa single pair of copper wires [25]. This technology is limited to a maximum range of1,000 to 4,500 feet (< 1 mile) from the central office, depending upon the speed.Modulation schemes proposed for VDSL include DMT and CAP. Moreover, asymmetrical version of VDSL is under study to operate in the range from 6.5 to 26Mbps [11].Table 2.3 presents a comparison of xDSL technologies along with FTTx and HFCstructures.2-18

Chapter 2Overview of current CATV network pTable 2.3 – Broadband access technologies.TechnologyBandwidth &MediumModulation Max. Distance ApplicationHFC(1988)42 Mbps down10 Mbps and possiblyto 30Mbps upfibre/coax cableQAM,QPSKVSB100 miles fibre trunk1 mile coax cableHigh-speed Internet,VoIP access, motionvideo, VoD, remoteLAN access, etc.FTTC(End 1980s)51 Mbps down andup fibre/coax/twistedpairTDM downTDMA up100 miles fibre trunk1 mile coax cable500 feet copperVideo broadcasting,HDTV Interactivemultimedia, ATMtrafficFTTH(End 1980s)155 Mbps down andupTDM downTDMA up100 milesonly fibre linksVideo broadcasting,HDTV Interactivemultimedia, ATMtrafficSuperPONs2.3 Gbps down311 Mbps upTDM downTDMA up60 miles Video broadcasting,HDTV Interactivemultimedia, ATMtrafficADSL(1995)1.5-8.2 Mbps down16-640 kbps upsingle twisted pairDMT3.4 miles at 1.5 Mbps3.0 miles at 2.0 Mbps2.3 miles at 6.3 Mbps1.7 miles at 8.2 MbpsTelephony, high-speedInternet access, motionvideo, VoD, remoteLAN access.G.Lite,DSL-Lite orSplitterlessASDL(1997)1.5 Mbps down512 kbps upsinge twisted pairDTM 3.4 miles Telephony, IP voiceservices over a singletwisted copper looppairHDSL(1991)1.5 Mbps duplex ontwo twisted-pairlines, up and down2.0 Mbps duplex onthree twisted-pairlines, up and down2B1QCAP2.3 miles T1/E1 service betweenserver and phonecompany or within acompany; WAN, LAN,server accessSDSL(1996)1.5 Mbps duplex, upand down2.0 Mbps on a singleduplex line up anddown2B1QCAP2.3 miles Same as for HDSLVDSL(1995)13 to 53 Mbps down,1.5 to 2.3 Mbps upsingle twisted pairDMT, CAP0.9 miles at 13 Mbps0.6 miles at 26 Mbps0.2 miles at 53 MbpsSame services asADSL + Videobroadcasting, HDTVInteractive multimedia,ATM traffic over Fiberto the Neighbourhood.2-19

Chapter 2Overview of current CATV network p2.8 ConclusionsThe increased need of Internet and broadband services has created an awareness of anddemand for high-speed access at mass-market prices. CATV networks evolved as analternative to satisfy these needs. However, the transition of CATV networks, fromanalogue broadcast unidirectional to a high-speed bi-directional digital medium,requires the reduction of noise and the availability of cable modem technology, whichwill provide bi-directional communication over the shared medium. The introduction offibre links in the cable plant, combined with progress in optical laser technology, hasmade possible the reduction of noise. Cable modem manufacturers first introducedproprietary equipment with limited capabilities in terms of multimedia applications andlow QoS support. Later, with the alliance of influential groups, such as the DVB-DAVIC-ETSI, MCNS-SCTE and IEEE 802.14 - ATM Forum, several cable modemstandards have emerged. However, only the DOCSIS protocol and the DVB/DAVICprotocol achieved the short time standardisation process demanded by Multiple ServiceOperators. Such protocols contribute to a wider acceptance of CATV networks.Telco’s, with its DSL technology, have entered the race for high-speed digital servicesover their point-to-point switched network. Although this technology (from telephonecompanies) is slightly more expensive than cable modem technology, DSL has theadvantage of a dedicated bandwidth for each node/service. Moreover, Fibre To TheCurb/Home is a promising technology, which can achieve very high data transmissionsin the order of 150 Mbps. Unfortunately, this technology is still in its early stages and atthe current time is too expensive for a residential or small business to afford.In the race for supplying multimedia broadband services and high-speed Internet accessto residential customers, CATV networks outdistance the competitors of digitalsubscriber line, fixed-point wireless, and fibre optic to the home systems. The race isongoing and for the foreseeable future it appears likely that cable and DSL technologywill lead the race to the home for supplying television, video on demand, homeshopping, video games, telephony, high quality video telephony, high-speed dataservices, and eventually hundreds of residential applications being envisioned or to beinvented.2-20

Chapter 3THE DVB/DAVIC PROTOCOL3.1 IntroductionDigital Video Broadcasting (DVB) has become one of the most exciting developmentsin the area of consumer electronics at the end of the twentieth century [88]. The DVBProject emerged from a group called the Launching Group of European broadcasters,consumer electronics manufacturers and radio regulatory bodies in 1992. Since then anumber of protocol specifications have emerged. This chapter presents a review of theDVB project and the main standards produced by this group. Special attention is paid tothe DVB/DAVIC protocol specification, which is the one that will be analysed,optimised and improved in subsequent chapters. Here the main characteristics, as wellas the architecture, MAC operation, and a description of the exponential backoffalgorithm and the splitting tree algorithm adopted by the DVB/DAVIC protocol aredescribed.3.2 DVB project overviewThe focal point of the European DVB project is the development of standards for thedelivery of digital TV over satellite DVB-S, terrestrial DVB-T and more recentlydelivery of digital data over cable links DVB-C for the downstream channel and DVB-RCC (or DVB/DAVIC) for the upstream channel or return channel.The DVB-S (EN 300 421 [29]) is a satellite specification designed to operate within arange of transponder bandwidths (26 MHz to 72 MHz). However, the cable network onthe downstream direction DVB-C (ETS 300 429 [31]) has the same core as the satellitesystem, but the modulation system is based on QAM rather than QPSK.3-1

Chapter 3The DVB/DAVIC protocol pInitially DVB recommendations did not cater for bi-directional communications.However, the implementation of interactive TV will require data in the reversedirection. Therefore, the DVB group is turning to other standardising bodies in order toproduce a specification with a wider range of applications, which will span beyonddigital TV broadcast such as high-speed Internet access, VoIP, video-telephony andvideoconferencing, among other services.At the initial stages DVB was evaluating which standard would be the most suitable forthe delivery of Motion Picture Experts Group (MPEG-2) audio/video streams overCATV networks. At its meeting in July 1997, DVB announced it would adopt theDAVIC 1.2 specification. Thereafter, both the UK and the European CableCommunication Associations (ECCA) announced that they were going to support theDVB/DAVIC standard.In order to accomplish compatibility with the other standards under development, DVBrequested from the IEEE 802.14 group to optionally include “DAVIC 1.2 Part 8” in it’sspecification, which dealt with lower layer protocols and physical interfaces for coaxialcable (including the MAC functionality). The European Telecommunication StandardETS 300 800 [32] (also known as DVB/DAVIC) has now been produced as the baselinespecification of the interaction channel for CATV distribution systems. Shortlythereafter, the European Telecommunication Standards Institute ETSI officiallyaccepted this standard, which became the European Standard “ETSI ES 200 800” [34],in April 2000.3.3 EuroModemMeanwhile the process of standardisation of the ETSI ES 200 800 protocolspecification, the European Cable Communication Association (ECCA) [131] andEuroCableLabs (ECL), as the centre of competence of the European cable operators,decided in January 1999 to push and co-ordinate the activities for the marketintroduction of the DVB/DAVIC compliant cable modems (based on the standards ETS300 800 and ETSI ES 200 800). Thus, the EuroModem project emerged and created thetechnical specification of an external cable modem that fulfils the requirements of the3-2

Chapter 3The DVB/DAVIC protocol pEuropean cable operators. The final EuroModem specification [36] was published afterapproval by the ECLs in May 1999.Two different types of EuroModem devices have been defined. The class “A”EuroModem is the basic version and is used mainly for high-speed Internet access. Thishas functionality similar to that of a DOCSIS 1.0 compliant modem. Using a class AEuroModem a secure data transmission is possible due to the defined encryptiontechnique. The class “B” EuroModem is the enhanced version supporting someadditional features. For instance, it is possible to deliver high-quality telephony servicesor Integrated Services Digital Network (ISDN) connections and also IP telephony(VoIP) with a guaranteed Quality-of-Service (QoS) level. A common telephonyinterface allows connection of a telephone device directly to the class B EuroModem.This modem has functionality similar to that of the DOCSIS 1.1 compliant cablemodem. For a further comparison of these three protocol specifications the readers arereferred to [84] and [109].3.4 DVB/DAVIC reference modelThe system reference model for interactive services for the DVB/DAVIC protocolspecification is shown in Figure 3.1. In the system model there are two channelsestablished between the service provider and the user.Broadcast ChannelBroadcastServiceProviderBroadcastNetworkAdaptorBroadcastDeliveryMediaBCSet Top Box (STB)/EuroModemBIMInteractive ChannelsIIMSTUInteractiveServiceProviderInteractiveNetworkAdaptorInteractionNetworkNIUEndUserInteractive ChannelNetwork IndependentNetwork DependentFigure 3.1 – DVB system reference model.Network Independent3-3

Chapter 3The DVB/DAVIC protocol pThey are the Broadcast Channel (BC) and the Interaction Channel (IC). The BC is aunidirectional broadband broadcast channel including video, audio and data. The IC is abi-directional interaction channel, which is established between the service provider andthe user for interaction purposes. It is formed by the following paths:• Return interaction path (also referred to as upstream channel): from the user to theservice provider. It is used to make requests to the service provider, transmit userdata or to answer questions.• Forward interaction path (also referred to as downstream channel): from the serviceprovider to the user. It is used to provide information generated by the serviceprovider for the user and any other required communication for interaction serviceprovision.Data connectivity is achieved via STB or EuroModem, which contain a NetworkInterface Unit (NIU) for interfacing with the cable network. The NIU consists of theBroadcast Interface Module (BIM) and the Interactive Interface Module (IIM). The userterminal provides an interface for both the broadcast and interaction channels. Theinterface between the user terminal and the interaction network is via the IIM.In the upstream channel, a 64-byte (upstream) slot structure is used and in thedownstream channel, the packet structure is based on 188-byte MPEG2 TS frames.Appendix B presents a description of packet formats, PDU structures, signallingmethods and protocol configurations.3.5 DVB/DAVIC MAC operationOne downstream channel can manage up to eight upstream channels by using MACcontrol information (referred here as MCI, see Figure B.4). This MCI field plays a vitalrole in the operation of this communication system, because it contains synchronisationinformation of the upstream slots. Its main functionality is to co-ordinate the usage,assign access modes, and indicate if reception of contention-based slots was successful.Each slot is assigned one of the following four classifications from the INA: ranging(for synchronisation and calibration purposes), contention (for light traffic load and3-4

Chapter 3The DVB/DAVIC protocol pMAC control message transmissions), reservation (for bursty or high traffic load) orfixed slots (for constant bit rate traffic).These MPEG2 frames with MCI information are transmitted in the downstream channel(at least) once every 3 ms when the upstream data rate is 6.176 Mbps [grade D], 3.088Mbps [grade C] or 1.544 Mbps [grade B], and every 6 ms for 256 kbps [grade A]). TheMCI field describes up to 36, 18, 9 and 3 upstream slots for grades D, C, B, and A,respectively. The limits between access regions allow the EuroModem (which is alsoreferred to as EM, NIU or station) to know when to send data on contention at a timethat risk of collision with data of reservation or fixed-rate regions does not exist.Previously in [101] and also in [34] the authors have reported that several functions areperformed by the MAC protocol for connection control and data transmission asdepicted in Figure 3.2. On power-on or reset, the initialisation and provisioningprocedure makes sure that an NIU is capable of tuning to the correct channel in theupstream and downstream directions and that it can receive the basic networkparameters.Initialisation and R egistration ProcessIN A / H eadendN IU / EuroM odemProvisioning Channel M essageD efault Configuration M essageInitialisation andprovisioning signallingSign-on M essageSign-on R esponse M essageR anging and Pow er Calibration M essageRanging and Power CalibrationResponse M essageInitialisation Com plete M essageSign-on, ranging andcalibration signallingC onnect M essageConnect Response M essageConnect ConfirmConnection signallingFigure 3.2 – DVB/DAVIC initialisation and registration process.3-5

Chapter 3The DVB/DAVIC protocol pThen, the sign-on, ranging and calibration are performed in order to adjust the internalclock and the transmission power of the NIU according to the specific transmissiondelay and cable attenuation. The initial connection is also established by default. TheMAC protocol also carries out the establishment and release of logical connections andallows readjustment of parameters, as well as performing an exchange of keys andestablishment of a secure connection. Here Diffie-Hellman and Data Encryption System(DES) are used.Once an NIU has initialised and registered with the INA, bandwidth reservation isprovided by both per-packet and per-session (or connection) basis. The former is knownas Reservation Access and is provided by sending reservation requests to the INA. Thefirst version of EuroModems (class A) will support this functionality as a mandatoryaccess mode. The latter is known as Fixed-rate access, where the INA provides either afinite amount of slots (in the fixed-rate access region) to a specific NIU or a given bitrate requested by an NIU until the INA stops the connection on NIU's demand. Thisaccess mode will be an optional functionality for EuroModem Class A, and mandatoryfor EuroModem Class B.3.5.1 Reservation access mode operationThe reservation-access mode with its dynamic slot-allocation feature is the main accessmode of the DVB MAC protocol for the transmission of data packets via the upstreamchannel. For this access mode, the DVB/DAVIC group has adopted two contentionresolution algorithms, which are used to resolve collisions: the exponential backoffalgorithm and splitting tree algorithm. The splitting tree algorithm takes advantage ofthe exponential backoff algorithm in the sense that feedback and allocation informationallow a station, (with new incoming arrivals) to compete for contention-based slotswithout risk of collision with backlogged stations. In addition, this algorithm makes useof minislots, which decreases the risk of collisions, since one contention-based slot isdivided into three minislots of 21-bytes long transferring shortened reservation requestmessages, increasing the probability of successful request transmissions andconsequently improving the system performance.3-6

Chapter 3The DVB/DAVIC protocol pThe reservation-access mode uses the following ‘Contention-Resolution-Grant Cycle’,(CRGC) for data transmissions, as shown in Figure 3.3:1) Wait until a ‘Reservation_ID’ has been received (which allows an NIU to send areservation request).2) When a packet arrives, send a reservation request message, using theReservation_ID assigned, in a contention-based slot.3) Wait until the INA sends the following MPEG2 frame with signalling information(MCI) and check in the reception indicator field whether the reservation wasreceived successfully.4) If the request resulted in a collision, use the CRA selected (exponential backoffalgorithm or the splitting tree algorithm) to retransmit the reservation request aslong as collisions are resolved.T im e fo r n e x tM P E G 2 F ra m ew ith M C I & G ra n tIn fo rm a tio nT im e fo r n e x tM P E G 2 F ra m ew ith M C I & G ra n tIn fo rm a tio nIN A (H E )R e s e rv a tio n _ IDA s s ig n m e n tM P E G 2 F ra m e(M C I)R e s e rv a tio nR e q u e s tM P E G 2 F ra m e(M C I/ R e s e rv a ito n G ra n t/F e e d b a c k )D a ta P a c k e t T ra n s m is s io nD a ta s lo t 1D a ta s lo t 2D a ta s lo t 3N IU (E M )P a c k e tA rriv e sT im e u n tilc o n te n tio n s lo tT im e u n til s u c c -fu l re q u e s t (in c a -s e o f c o llis io n s )T im e u n tilre q u e s t isg ra n te d& p a c k e tre c e iv e dWaiting delay Contention delay Grant delayD a ta s lo t nFigure 3.3 – Contention-Resolution-Grant Cycle.3-7

Chapter 3The DVB/DAVIC protocol p5) If the request was transmitted successfully, wait until the INA grants the request totransmit the data packet.6) Finally, after transmitting the current data packet in the reserved slots assigned bythe INA, if there are more data packets to transmit repeat the process starting in 1).In the following sections, a description of the operation and dynamic that eachcontention resolution algorithm uses to resolve collisions is presented.3.5.2 Exponential backoff algorithm overviewIn this sub-section we present a review of the operation of the exponential backoffalgorithm adopted by the DVB/DAVIC protocol. Furthermore, in Section 6.3 weapproach a performance analysis and optimisation of this algorithm.When an NIU wants to transmit a reservation request in a contention slot, it choosesarbitrarily one contention slot in the next group of contention slots described by thesignalling information field, as depicted in Figure 3.4. In case of collision, thisalgorithm defines how many cells a station needs to let pass before it can transmit. Thisnumber of cells is computed as a uniform random integer variable in the range of [0-2 backoff ].P k A r r iv a lS t a r t C o n t e n t io nP r o c e s sT x . R e q . i n a R N D C SC SR e qT xR N D= C n t e n t i o n S l o t= R e q u e s t= T r a n s m it= R a n d o mi = in it ia l b a c k o f f e x p .t = t r u n c a t e d b a c k o f f e x p .C o lli s i o n ?N OS u c c e s s f u l R e q . T xW a it f o r g r a n t a n d t h e nt r a n s m it d a t a s l o t sY E Si = i + 1 , u n t il i < = tb = R N D ( 2i)b a c k o f f ( b - 1 ) C S &T x . R e q . I n t h e b t h C SE n dFigure 3.4 – Block diagram of the exponential backoff algorithm.3-8

Chapter 3The DVB/DAVIC protocol pThe backoff parameter is first initialised with an initial backoff exponent (also termed as‘Minimum Backoff Exponent’). This parameter is updated according to the receptionindicator received from transmission. The truncated backoff parameter (also referred toas ‘Maximum Backoff Exponent’) is the maximum allowable backoff. If a collisionoccurs, the backoff value is incremented by one. Once the backoff value reaches themaximum number determined by the truncated backoff, the backoff remains at thisvalue regardless of the number of subsequent collisions. After every successfultransmission, the backoff value is reset to the initial backoff value. The values of initialand truncated backoff are sent by the INA to the stations at the beginning of theconnection and remain the same until the NIUs stop the connection with the INA. Thesetwo values are equal for all the NIUs as opposed to the backoff value that is different ateach station.3.5.3 Splitting tree algorithm overviewAs defined in Appendix B.2, minislots are only used to transmit reservation requestwhen this algorithm is used. After an NIU sends a reservation request in a contentionbasedminislot, it waits for feedback information from the INA. In the case of collision,the NIU enters the contention resolution cycle and activates its resolution state.3-9

Chapter 3The DVB/DAVIC protocol pThe resolution is carried out according to an INA controlled splitting tree algorithm asshown in Figure 3.5. All necessary information to resolve collisions is provided in theReservation Grant Message, which contains minislot feedback (such as ‘FeedbackOffset’ and ‘Feedback Collision Number’ 1, 2 and 3) and minislot allocationinformation (Such as ‘Stack Entry’, ‘Entry Spreading’, ‘Number of Allocations’,‘Allocation Offset’ and ‘Allocation Collision Number’).In the current Reservation Grant Message, If ‘Stack Entry’ is not set, an NIU may enterthe contention process only when the ‘Allocation Collision Number’ is equal to zero. If‘Stack Entry’ is set, the NIU may enter the contention resolution in any of thecontention-based minislots, independent of the value of ‘Allocation Collision Number’.In both cases the random number for the minislot selection in the range between 0 and‘Entry Spreading’ should be in the window from 0 to 2 before sending the request.StartContentionProcessDoIf(Entry_Stack = 0)wait for group of minislots withAllocation_Collision_Number = 0Elsewait for group of minislotsR = Random(Entry_Spreading)Until R < 3Transmit request in minislot number Rwait for feedback/reservation grant= 0xFESuccessfulTransmissionFeedback_Collision_Number= 0xFFError/empty slotExit< 0xFFCollisionwait for group of minislot withAllocation_Collision_Number = Feedback_Collision_Numbertransmit request in minislot number random(3)Figure 3.5 – Block diagram of the splitting tree algorithm.3-10

Chapter 3The DVB/DAVIC protocol pThe ‘Feedback Collision Number’ is equal to 0xFF or 0xFE for idle and successfultransmission, respectively.All other values of the ‘Collision Number’ are numbered as collisions and used to selectthe retransmission minislots. The NIU should retransmit in a minislot having an‘Allocation Collision Number’ equal to ‘Collision Number’. The retransmission of thecollided request takes place in a minislot that is randomly selected from among thegroup of three minislots with the corresponding ‘Allocation Collision Number’3.5.4 Upstream slot structureIn Section 3.4, we have stated that the upstream channel is divided into discrete basicslots called upstream slots. A fixed number of upstream slots are grouped to form asignalling frame. A generic slot structure is shown in Figure 3.6 (for a 3.088 Mbpsupstream channel), where the INA determines the slot structure format (and reservationcapacity) by setting the number of ranging, contention, reservation and fixed slots ineach signalling frame, an example of which is depicted in Figure 3.6a. They are used forthe following purposes:• Ranging slots: These slots are used in the sign-on, ranging and calibrationsignalling process to measure and adjust the time delay and the power.Signalling Slot Structure3 milisecond periodSlot Configuration Field 1 Slot Configuration Field 2a)b)c)1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18Ranging slotContention slotReservation slotFixed slotMinislotFigure 3.6 – Slot structure.3-11

Chapter 3The DVB/DAVIC protocol p• Contention slots: are accessible for every station to send reservation requestscollisions are possible and solved by a contention resolution algorithm. Theseslots can also be used to send user data on low traffic loads, if this is enabled bythe INA.• Reservation slots: the user sends control information announcing his demandfor transmission capacity. He gets grants for the use of slots.• Fixed slots: are used for CBR traffic, these slots provides the best results in QoSregarding delays, jitters and loss of upstream data units.The slot structure is based on two Slot Configuration Fields (SCF). Each SCF describes9 slots in the order presented in Figure 3.6a (ranging, contention, reservation and fixedslots). The number of ranging slots transmitted during a long period of time (e.g. 60seconds) is insignificant, compared to the number of contention, reservation and fixedslots transmitted, and the performance of the network does not change if the rangingslots are not used at all. Thus, the use of ranging slots will not be considered in thesignalling frame structure for performance analysis.Furthermore, since fixed slots may or may not be supported in the first version ofEuroModem (class A), the performance analysis to be addressed in the followingchapters will be based mainly on signalling slot structures that consist of contention andreservation slots, as depicted in Figure 3.6b. In this structure the contention andreservation slots have been rearranged so that all contention slots are described in thefirst SCF, and in the second SCF when SCF1 describes only contention slots. This newarrangement decreases slightly the processing times of the INA and NIUs.As pointed out in Section 3.5.3, if the splitting tree algorithm is used to resolvecollisions, one contention-based slot is divided into three independent minislots thatcarry shortened request messages, as depicted in Figure 3.6c. When this algorithm isused, the number of contention slots described in the first SCF (or SCFs) can bedecreased.In general, the slot structure presented in Figures 3.6b and 3.6c will be used when theexponential backoff algorithm and the splitting tree algorithm are selected, respectively.3-12

Chapter 3The DVB/DAVIC protocol p3.6 ConclusionsAn overview of the evolution of the European DVB/DAVIC Cable CommunicationSystem has been provided. In the early stages of the cable data networks, the DVBgroup in collaboration with the European Cable Communication Association brought asignificant value for the broadcast data-services organisations by providing the set ofspecifications for the return channel (ETS 300 800, ETS 200 300, EuroModem,EuroBox) that will enable cable network operators to provide the next generation ofinteractive services.In this chapter we have also presented the network architecture and the majorconstituent components of the DVB/DAVIC protocol, which represent the mostimportant elements that describe this standard. Here the basic operation at the physicallayer and the Media Access Control layer were described along with the functionality ofthe contention resolution algorithms adopted.The definition of the MAC layer was important, since spectrum of cable networksavailable for upstream transmissions from subscribers to the headend is scare. As a finalremark, the DVB/DAVIC protocol fits optimally into a cable environment and providesan integrated solution for full-interactive service provision.3-13

Chapter 4SIMULATION AND ANALYTICALMODELLING OF DVB/DAVIC4.1 IntroductionAs reviewed earlier in Section 1.2.3.2, the DVB/DAVIC protocol requiresapproximations of both the exponential backoff algorithm and the splitting treealgorithm, which cannot be accurately modelled using existing analytical models due tothe variable bandwidth assigned to the contention and reservation access regions from(MCI) cycle to cycle. Therefore, the high degree of complexity and the need foraccurate results indicate the use of simulation techniques for the performance evaluationand optimisation of the DVB/DAVIC protocol. The use of a simulation model willallow us to examine a wide range of configurations needed for drawing generalconclusions, and to test new enhanced mechanisms that would increase the performanceof the DVB/DAVIC protocol in a relatively short amount of time.In this context, the aim of this chapter is to present both the simulation model developedfor the DVB/DAVIC protocol and the analytical model formulated for the validation ofthe simulation results.4.2 Simulation modelling toolsAccording to [70] there are three main simulation languages oriented to the modellingof communications networks. The first language is the Block Oriented NetworksSimulator (BONeS) DESIGNER, which is a graphically-oriented, general-purposesimulation language that contains many features for modelling communicationsnetworks.4-1

Chapter 4Simulation and analytical modelling pThe major data blocks are data structures and block diagrams. In order to build a modelin BONeS, one first defines a data structure that corresponds roughly to a message andits associated data field. The user then develops a block diagram that describes how thedata structures flow through the networks.The second language is SES/workbench, which is also a graphically-oriented, generalpurposesimulation language that contains many features for modelling computersystems and communications networks. The major building blocks are nodes, arcs andtransactions. To build a model in workbench, one defines a transaction that correspondsto a message. The user then develops a directed graph consisting of nodes and arcs,which describes how transactions flow through the network.The last simulation language is the OPtimised Network Engineering Tool (OPNET)Modeller. This software is a communication simulation language specially orientedtoward the modelling of communications networks that uses etwork, odes, andProcess Editors to build a simulation model.From these three simulation languages, OPNET Modeller (v6.0 [79]) has been used asthe simulation tool for the performance characterisation and optimisations of theDVB/DAVIC protocol because this simulation package contains an extensive set offeatures designed to support network modelling and provides an increased flexibility todevelop detailed custom models.4-2

Chapter 4Simulation and analytical modelling p4.2.1 OPNET modellingIn order to have an insight of how simulation models are implemented in OPNET, inthis section an overview of the hierarchical design used by this simulation package ispresented. OPNET models are based on a hierarchical three-level structure as illustratedin Figure 4.1.• etwork domain: The top-most level is the etwork domain (see Figure 4.1a). Therole of this domain is to define the topology of the communication network, thecommunication entities called nodes and their interconnection (using bus, point-topointor radio links). Based on these basic building blocks, more complex modelscan be developed.• ode domain: The next level is called the ode domain (Figure 4.1b). This leveldefines the functionality of each communication device that can be deployed andinterconnected to the network, (e.g. routers, bridges, terminals, switches, etc). Eachnode consists of traffic source generators, processors, queues and varioustransmitters or receivers allowing a node to be attached to communication links inthe network.a ) N e tw o rk d o m a inT e rm in a lT e rm in a lS e rve rP ro to - C C o d ein e a ch sta teS e rve rB rid g e /R o u te r/S w itc hW o rk sta tio nc ) P ro c e s s d o m a in (F S M )b ) N o d e d o m a ind e fa u ltF ra m e _ W a itin ginitS in kp ro c es sorT ra fficg e n e ra to rfrm _ wdef_ o ffS p a c in g _ E la p s e dR e ceiverM a cQ u e u eT ra ns m itte rF ra m e _ W a itin g _ L o w & &! C h a n n e l_ B u s ybsy_w! F ra m e _ W a itin g & &!C h a n n e l_ B u syd e fa u ltd e fa u ltspacingB u s y _ L o wfree_ wD e fe r p ro c e ssorF ra m e _ W a itin g _ L o wB u s y _ H ig hd ef_o nd e fa u ltFigure 4.1 – Stage diagram of OPNET simulation models.4-3

Chapter 4Simulation and analytical modelling p• Traffic generators are built-in objects used for simulating traffic sources. Thegenerated traffic stream has packet length and a packet inter-arrival distribution,which is selected from a set of supported probability distributions. Queues andProcessors are objects and are highly specifiable. Queues are used to simulatebuffers of communications links and protocol behaviour. Example statistics that aqueue can collect are current/average queue size, queuing delay and queue overflow.Each queue object might spawn multiple sub-queues in which packets can be stored.Processors have the capability of packet manipulation but no ability to store packetsand can be used as packet sinks or as traffic generators, when the traffic streamscannot be modelled using the predefined probability distribution.• Process domain: The last level is called the Process domain (Figure 4.1c). At thislevel the functionality of each queue or processor object is defined using a FiniteState Machine (FSM). States and transitions graphically define the progression of aprocess in response to events. In general, each FSM can define private statevariables and can make calls to code in user provided libraries. FSM are dynamicand can be spawned by other FSMs during simulation in response to specific events.Dynamics of FSM dramatically simplify specification of protocol that manages ascalable number of resources or sessions. Finally, each state of a process modelcontains ‘Proto-C’ code, supported by an extensive library of functions designed forprotocol programming.4.2.2 Common Simulation Framework (CSF)Although OPNET models are accompanied by an extensive library of predefinedcommunications protocols covering all seven layers of the Open System Interconnection(OSI) Reference Model, no MAC protocols are available that are suitable for modellingbi-directional CATV networks. Thus, the high demand for evaluating emergingprotocols such as the IEEE 802.14 and DOCSIS, have motivated MIL3 (now OPNETTechnologies Inc.) and Cablelabs to develop the Common Simulation Framework(CSF) [77], which provides the basic building blocks of a typical CATV network fordelivering data applications without defining the functionality for the MAC or layersabove.4-4

Chapter 4Simulation and analytical modelling pThis framework not only includes statistics collection models and interfacing with othernetwork technologies (e.g. Ethernet, ATM, etc), but also the appropriate links betweenthe Headend (HE) equipment and cable modems for upstream and downstreamtransmissions. The latest version of the CSF (v.13) covers the basic functionality of theMAC and PHY layers of a DOCSIS v.1.0 compliant cable modem and Cable ModemTerminations System (CMTS). Some additional functions, such as prioritisation [102],committed information rate scheduler algorithm [120] and two schemes to reduce thenetwork's recovery time [105] and [106] were added at the University of Sheffield.The network architecture of the generic CSF model is presented in Figure 4.2. Itconsists of the Headend node, Statistics Collection block, Network cloud, and CableModems.a) Headend: The main purpose of this node is to receive messages from the upstreamchannel and to relay them to either the downstream channel or the Network cloud. Atraffic source module within the Headend node models incoming traffic from theNetwork side interface. Additionally, some other functions performed by this node(at the MAC layer) are CM initialisation, registration, scheduling of upstreamtransmission and routing of incoming frames to the Statistics Collection module.b) Cable Modem: This node connects both the upstream and downstreamunidirectional channels. With the use of the specific MAC protocol, it generatestraffic that is sent to the upstream channel. The CSF provides different types ofsource traffic, such as ON-OFF exponential distributed bursty traffic, 53-byteisochronous streams and World Wide Web (WWW) traffic flows.Upstream ChannelStatisticsCollectionInternet/NetworkcloudHeadendCableModemsDownstream ChannelFigure 4.2 – Generic CSF model.4-5

Chapter 4Simulation and analytical modelling pEach CM can have more than one type of traffic source active as well as more thanone source of the same type.c) Upstream and Downstream channels: These channels model the pair offrequencies allocated to a set of CMs for bi-directional communications with eachother and the backbone network/internet. The data rates and capacity of these linksare parameters defined by the specific protocol and can be either symmetrical orasymmetrical. There are additional link level statistics provided, such as throughput(which includes protocol overhead), utilisation, collision events, multiplicity etc.d) Statistics Collection: This module interfaces with the HE node and collects framesreceived from the upstream channel in order to produce performance analysisstatistics, such as mean access packet delay, system throughput, jitter, utilisation,global offered load, etc.e) etwork Cloud: This module interfaces with the HE node and is used to modelcommunications and traffic load originating from a backbone, which is not part ofthe CATV network. The generic CSF does not include any sample architecture forthis module.4.3 DVB/DAVIC simulation modelThe simulation model for the DVB/DAVIC protocol is based on the main networktopology of the CSF (v.13). All the functionality of the MAC and PHY layers of theDVB/DAVIC protocol (including the INA and NIUs) have been developed andincorporated to the CSF, replacing the functionality of the MAC and PHY layers of theDOCSIS protocol, respectively. Suitable changes were also applied to the StatisticsCollection module to support new performance statistics. Furthermore, one trafficsource generator was added to this model in order to emulate Internet traffic as proposedby the IEEE 802.14 Working Group [53].4-6

Chapter 4Simulation and analytical modelling p4.3.1 DVB/DAVIC network descriptionThe DVB/DAVIC simulation model uses a similar network topology as proposed byNarayanaswamy [77] for the CSF. This network topology is presented in Figure 4.3a. Itconsists of two logical buses (upstream and downstream) and Network Interface Units(subscribers) which are connected to the upstream and downstream channels tocommunicate with the Interactive Network Adaptor (HE). It also contains an INAmodule, which processes upstream data, evaluates upstream requests and sends data andsignalling information to all NIUs listening downstream.The corresponding representation into an OPNET model is as illustrated in Figure 4.3b.Several network models were created containing different number of nodes (rangingfrom 6 to 700 nodes), according to the performance analysis addressed.a) Network Topology b) OPNET Network ModelDownstream channelStatsCollectionINA NIU NIU NIUn1n6INAUpstream channelNetwork CloudFigure 4.3 – Network topology.4-7

Chapter 4Simulation and analytical modelling p4.3.2 Network Interface Unit designThe interdependent components of the NIU system consist of the traffic sources, atraffic sink (to record statistics), the IU MAC and transceivers, as shown inFigure 4.4a. The traffic sources generate packets that are delivered to the IU MACusing a packet stream connection. This is represented by the Packet_In link in thefigure. The IU MAC transfers and receives packets from the transmitter/receiversystem. These actions are represented by the Packet_Out and Packet_In connections inFigure 4.4a, respectively. All user data coming from the downstream channel isforwarded to the traffic sink for collecting statistics using the Packet_Out connection.a) NIU Design b) NIU Representation in OPNETPacket-InPacket-OutTrafficSourcesTrafficSinkNIU MACIUManagerIUStationChildTransmitterSystemReceiverSystemsrc_type_1bus_receiversrc_type_2bus_transmitterNIU_MACPacket-OutPacket-Indownstream_sinkanimationFigure 4.4 – NIU node description.The NIU representation into a node model in ONPET is as depicted in Figure 4.4b. Twotraffic sources were used to emulate Internet traffic and isochronous streams• Traffic Source 1: This traffic source emulates Internet traffic according to a customProbability Density Function (PDF), as recommended by the IEEE 802.14 WorkingGroup [53]. The distribution message size is shown in Figure 4.5.Probability0.60.060.04 0.020.250.0364 124 256 512 1024 1518Message size (bytes)Figure 4.5 – Internet packet distribution.4-8

Chapter 4Simulation and analytical modelling pThe average size is ≈ 368 bytes or 8.3 ATM cells and the message inter-arrival timeis exponential distributed with mean T = 1/λ, where λ varies according to the offeredload (e.g. 32 or 64 kbps). The minimum packet size of 64 bytes corresponds to anacknowledgement packet that is exchanged during a TCP session and the packet sizeof 1518 bytes corresponds to the maximum size of an Ethernet packet that is mainlyFigure 4.6 – NIU Manager process – FSM.used in PC Networks.• Traffic Source 2: This traffic type emulates a CBR stream. It is a modifiedimplementation of the ON-OFF source provided by the CSF, which is used togenerate isochronous streams. All sources create packets of constant length, variablebetween simulation runs from 20 to 1518 bytes to cover a large range of applicationswith a constant inter-arrival rate. The time, t, that each source starts generatingpackets was exponentially distributed with a mean of 1 second. This prevents NIUsfrom issuing the first reservation request simultaneously and causing an excessivenumber of collisions at the beginning of the simulation.The most important block in the NIU node is the IU MAC (Figure 4.4a). This block isbased on two asynchronous processes: The IU Manager process handles upstream/downstream traffic and performs the initialisation and registration process depicted inFigure 3.2.4-9

Chapter 4Simulation and analytical modelling pThe finite state machine used for the IU Manger process is as illustrated in Figure 4.6.Packets from the traffic source generators and signalling frames that describe the use ofthe upstream channel are transferred to the IU Child process (see Figure 4.7). TheChild process performs bandwidth request, collision detection/resolution and packettransmissions.Figure 4.7 – NIU Child process – FSM.4-10

Chapter 4Simulation and analytical modelling pEach state of these process models contains ‘Proto-C’ code, which is supported by anextensive library of functions designed for protocol programming. An example of whichis presented in Figure 4.8 for the state “Transmit opportunity”.Figure 4.8 – Proto-C code of the state: Tx. Opportunity.4-11

Chapter 4Simulation and analytical modelling p4.3.3 Interactive Network Adaptor designThe Interactive Network Adapter (INA) node consists of the IA MAC and four linkaccess points to interact to the CATV network, the Network Cloud and StatisticsCollection node as pointed out in Figure 4.9a.The representation of the INA into an OPNET node is shown in Figure 4.9b. Here theIA MAC was connected to the downstream bus access point via a statistical wire, thuspreventing the IA MAC sending another frame to the bus. Currently, the trafficgenerator (Downstream source) has been disabled, thus no data packet would betransmitted downstream, since the main research focuses on the performance analysis ofthe upstream channel, which is the only critical network resource because of thereservation access mechanism and its marked asymmetry. In addition, the upstreamscheduler is part of the IA Manager process.The IA MAC node is also modelled as two asynchronous processes, the IA Managerand the IA Transmitter (TX) process. The IA Manager process handles the MACInitialisation (which consists of initialisation, provisioning, ranging, calibration andconnection signalling), receipt of upstream packets and downstream traffic. The IA TxController handles transmission of data and MAC signalling messages generated by theIA Manager process.a) INA Designb) INA Representation in OPNETPacket-OutPacket-OutStatisticsCollectionINA MACIAManagerTransmitterSystemfrom_network_cloud_txto_network_cloud_txto_stat_collection_txDowstreamTrafficIA TxControllerReceiverSystemINA_MACDownstream_sourcePacket-InPacket-Infrom_bus_rxto_bus_txFigure 4.9 – INA node description.4-12

Chapter 4Simulation and analytical modelling pThe finite state machine of the IU Manager process is depicted in Figure 4.10. Theoperation of this state machine is as follows. Upon simulation start the state variablesare initialised. The MAC Initialisation procedure makes sure that an NIU cansynchronise with the INA and that it can receive the basic network parameters. When allNIUs have established a connection with the INA, an initial frame containing signallinginformation is sent to the cable network.Figure 4.10 – INA Manager process – FSM.The state machine then switches from the idle state to one of the five other possiblestates (that were considered for this study), representing the following events: 1) datapacket arrival from the upstream channel, 2) resource request message arrival from theupstream channel, 3) reservation request message arrival from the upstream channel, 4)MCI timer expired, and, 5) data packet arrival from the traffic generator. For each ofthese events, the appropriate MAC message or data packet is created and passed to theIA TX Controller process for further transmission on the downstream channel.4-13

Chapter 4Simulation and analytical modelling pThe IA TX Controller is a spawned state machine (see Figure 4.11) or child process tothe INA Manager. Its purpose is to receive either MAC messages or data messages fromits parent and transmit them on the downstream channel as MPEG2 frames. The IA TXController process interacts with the bus object (to_bus_tx) via a statistical wire thatinforms when the bus is free (Figure 4.9b). When the downstream bus is idle and thereare MAC messages or data packets to transmit, the IA TX Controller serves first theMAC messages in a First In First Out (FIFO) order, and assembles up to three MACmessages into an MPEG2 frame, which is then delivered to the downstream channel.Figure 4.11 – INA TX Controller - FSM.4.4 Analytical modellingThe analytical evaluation of CATV protocols, and by implication DVB/DAVIC, provesvery difficult due to the increased complexity involved in such protocols, as reviewed inSection 1.2.3.2. The downstream channel, which is a unidirectional broadcast channelused only by the INA, can be easily modelled as a simple delay queue with a constantservice rate. However, the modelling of the upstream channel is complicated due tohybrid multiple random access-reservation MAC.4-14

Chapter 4Simulation and analytical modelling pA DVB/DAVIC upstream channel can be modelled as a triplet of virtual channels asdepicted in Figure 4.12. A portion of the upstream channel is dedicated to reservationand fixed-rate access in the reservation channel (R-channel) and the fixed channel (Fchannel)respectively, while reservation requests are placed in the contention channel(C-channel). There is a strong correlation between the traffic patterns in the three virtualchannels. This is because traffic in the C-channel accounts for reservation/resourcerequest, which if successful, trigger transmissions in the R or F channels.According to [10] it is possible to derive simple analytical models for such a systemusing the C-channel and considering the R and F channels as a single virtual channel.The throughput derived (S) for the generalised multi-access reservation system, inaccordance with [10] results in:1S =v1+Sc(4.1)where v is the No. of time units required to transmit one request when a data packetrequires one time unit, and S c is the normalised throughput of the C-channel. Thismethod requires that the throughput of the C-channel can be calculated.CRUpstream ChannelFRandom AccessContention-based region TDMA-Reservation Access Fixed-rate AccessContention Reserved Fixed Contention Reserved FixedUpstreamContention SlotReserved SlotFixed-rate SlotFigure 4.12 – Upstream channel model with virtual contention reservation andfixed-rate channels.4-15

Chapter 4Simulation and analytical modelling pFor certain shared access methods such as the Slotted Aloha, splitting tree or TDMA themaximum normalised throughput is 1/e. 0.478 and 1 respectively [10].At the time of writing this dissertation no accurate analytical model for CATVprotocols, which deal with variable reservation access (queues with contention), hadbeen proposed. When modelling the hybrid DVB/DAVIC protocol, two areas requirespecial consideration:• Analysis of the contention resolution algorithm. We have focused here on thestability of contention resolution algorithms and the time to resolve collisions.• Variable bandwidth of the C-channel. This effect is a direct consequence on theabove feature and the CRA adopted. Depending on the offered load,⎡ MCI⎤CSsc ∈ ⎢ ⋅CC,CC⎥⎢⎣MCImax_slots ⎥⎦(4.2)where c is the bandwidth of the C-channel, MCI CSs is the minimum number ofcontention slots described per MCI frame, and MCI max_slots is the maximum numberof slots described in each MCI frame. The upper bound is realised when the cablenetwork is idle and every MCI frame consists only of contention slots, thus c = CC.The lower bound is realised when the cable network is in congestion and thebandwidth of the C-channel becomes the minimum number of contention slots perMCI frame times the number of MCI frames per second. Thus, the bandwidth of theR and F channels becomes,r + f = MCImax_−(4.3)slotsMCI CSsBy implication, the bandwidth of the R and F channels varies and is correlated to c.This variation is a function of the offered load and the efficiency of the contentionresolution algorithm, which specifies the number of successful transmissions in theC-channel. Therefore, even if the C-Channel was approximated by the SlottedAloha of TDM access algorithm, the throughput calculation proves problematic, asthe channel bandwidth is variable.4-16

Chapter 4Simulation and analytical modelling pConsidering the above limitation, the formulation of an accurate analytical modelbecomes highly complex. Despite the high level of complexity associated with theproduction of an analytical model, a form of validation process is essential, even of thebasic functionality of the DVB/DAVIC protocol in order to verify correctness of thesimulation model.The analytical model formalised in Sections 4.4.1 to 4.4.3, address some questions thatconcern performance evaluation of a new protocol, (i.e. the knowledge of packet accessdelays and maximum system throughput achieved per stations). This is one of the firstissues that participant vendors and cable network manufacturers have to address, asthere is a significant delay associated with transmissions in the upstream channel thathas to do with the CRGC, even where there are not contending users in the cablenetwork. The analytical model presented in Section 4.4.1 and Section 4.4.2 is based inthe following reasonable assumptions:• The bandwidth of the F-channel for fixed-rate access is not considered. This is anadvanced functionality included in the second version of EuroModems [36].• In accordance with the DVB/PDAVIC protocol, at least one MCI frame should betransmitted in the 3 ms period. We consider that the DVB/DAVIC transmits 4 MCIframes per 3 ms period. Thus the number of slots described per MCI frames is of 9slots, with a 6.17 Mbps upstream channel.The following formulation addresses issues of mean packet access delay and maximumsystem throughput achieved for a single upstream channel. The quantities andabbreviations that will be used throughout the analysis have been defined in Table 4.1.4-17

Chapter 4Simulation and analytical modelling pTable 4.1 – Symbols and abbreviations.Symbol Description Value or range if applicableλ Packet arrival rate in Pk/s (also referred to as offered load in kbps)µ Average number of packets transmitted per second -XMean packet service time -ρ Utilisation factor -X idle Service time when the system is idle -X busy Service time when the system is busy -2σXVariance of the mean service time 0L Mean number of packets in the system (including the packet in service) -Pk size Length of the packet to be transmitted in bytes [64 –1518]Pk slot Number of slots required to transmit a packet -Pk mci Number of MCI frames required to transmit a data packetAAL5 header Protocol overhead added because of the ATM-AAL5 encapsulation 8 bytesAAL5 PDU Length of the ATM-AAL5 Packet Data Unit 48 bytesWDMean average waiting time -Mean access end-to-end delay from the NIU to the HE -eteMCI pk Number of MCI frames required to transmit a packet -MCI max_slots Maximum number of slots described in each MCI frame 9 slotsMCI CSs Minimum number of contention slots described per MCI frame 1 slotMCI t MAC Control Information interval 0.750 msD prop Propagation delay from the NIU to the HE 0.051 msD sl_tx Slot transmission delay 0.083 msCC Upstream channel capacity in Mbps 6.176P 0P 1TTTLippbbcbcD lahD tx_MPEGD interleaveD IU_procDS_sizeProbability that the system is in an idle stateProbability that the system is in a busy stateAverage time of the idle periodAverage time of the busy periodAverage time of the busy cycleAverage packets in queue per busy cycleScheduler look-ahead delayDownstream transmission delay of an MPEG2 TS frameDelay incurred to enable the correction of burst noise induced errors.NIU processing delayTotal size of the MPEG2 TS frame to be transmitted downstream includingFEC information.204 bytesD ws_bit_rate Downstream data rate 42 MbpsL xSS max_large_packetsS max_exp_backoffS max_split_treeAverage number of packets served per secondThroughputMaximum system throughput for large packet sizesMax. throughput for different packet sizes using the exp. backoff algorithmMax. throughput for different packet sizes using the splitting tree algorithm4-18

Chapter 4Simulation and analytical modelling p4.4.1 Mean access delay formulation for a single node scenarioFor the upstream direction, the Reservation Access mechanism of the DVB/DAVICMAC protocol can be modelled as a queuing system. The upstream timingrepresentation, depicted in Figure 4.13, permits a graphical view of the dynamic of ourqueuing system. This particular figure is shown for a first-come-first-serve orderservice. In this timing diagram the horizontal line just below the upstream channelrepresents the queue (at the NIU) and the horizontal line above the channel representsthe service facility. An arrow approaching the queue from below indicates that anarrival of a data packet has occurred (represented by A pk ). Conversely, arrowsemanating from the server indicate the departure of a data packet from the queue (D pk ).In this queuing system there are mainly two types of delays involved in the transmissionof a data packet, namely waiting delay in queue (W) and service delay (X). Additionally,when a packet departs from the server, there is a constant delay involved, which iscaused by the transmission delay and the propagation delay before the data packetreaches its final destination. Here, only the transmission delay of the last slot of thecurrent data packet is only considered. The transmission delay of the other data slotsform part of the service delay. Hence, the general equation to calculate the total meanaccess end-to-end delay,D ete , is given by:D ete X + W + DSlTx+=_D(4.4)propwhere X and Ware the mean of service and waiting delay, respectively.MCI 0MCI 1 MCI 2MCI 3MCI 4MCI 5MCI 6MCI 7MCI 8MCI 9ServerUpstreamChannelQueueD Pk0D Pk1D Pk2R D D R D DRDDW 1W 2A Pk0A Pk1A Pk2X 1X 2X 0Contention SlotRContention Slot withReservation RequestDData SlotFigure 4.13 – Upstream timing diagram.4-19

Chapter 4Simulation and analytical modelling pThe transmission delay of a data slot and the propagation delay are given by Equations4.5 and 4.6, respectively.D512= 83 s(4.5)CCSl _ Tx= µlD prop= = 50µs(4.6)0.67 ⋅ cIn Equation 4.5, CC corresponds to the upstream channel capacity set to 6.176 Mbpsand l in Equation 4.6 corresponds to the maximum length of the HCF network, from theINA to the further NIU. For this analysis l was set to 10Km and c is the speed of lightconstant (≈3×10 8 m/s).The solution of this queuing system depends on the arrival and service distribution ofthe data packets. Assuming that the data packets follow a Poisson distribution (e.gexponential distributed inter-arrival times) and by obtaining the mean service time ( X )of Figure 4.13, we can approach the solution by using an M/G/1 queuing system. Suchsystem is characterised by a Poisson arrival process (at a mean rate of λ arrivals persecond with a mean inter-arrival rate of 1/λ) and with a general service time distribution(with a mean service time of X seconds).According to [45], a well-known result for the M/G/1 system is given by the Pollaczek-Khintchine (PK) formula presented by Equation 4.7.2 2 2ρ + λ ⋅σXL = ρ +(4.7)2 ⋅ (1 − ρ)This formula gives the average number of packets in the system L, where ρ is theutilisation factor given byρµλ= = λ ⋅ X , ρ < 1, andσ 2 is the variance of the service-Xtime distribution. From this formula the expected waiting time W in queue can beobtained by using the well-known Little’s theorem, L = λ·T [10], where T is the meanwaiting time in the system given by:T = X + W(4.8)4-20

Chapter 4Simulation and analytical modelling pThus, the waiting time in the queue results in:Then, by substituting the utilisation factor2 2 2ρ + λ ⋅σXρ +2 2 22 ⋅ (1 − ρ)ρ + λ ⋅σXW = − X =(4.9)λ2 ⋅ λ ⋅ (1 − ρ)ρ = λ ⋅ Xformula for the mean waiting time in the system, as:The mean and the variance of the service time ( X and2in Equation 4.9, we can obtain the2λ ⋅ X + λ ⋅σXW = (4.10)2⋅(1− ρ)2σ respectively) are neededXalso. From analysing Figure 4.13, we can see that the mean service time has threevalues, represented by X 0 , X 1 and X 2 . A service time of X 0 is given to packets that findthe system idle. In other words, the packets start being processed immediately and donot have to wait in queue to be served. A service time of X 1 is given to packets that findthe system busy and have to wait in queue before they are processed.The service time X 2 (=X 1 ) is for special cases when packets find the system idle andhave to wait for a very short period of time before being processed. This is because suchpackets arrive at the end of the current MCI frame and the DVB/DAVIC reservationmechanism is unable to start service immediately, due to the time required for thepropagation and the slot transmission delay. In other words, when a packet arriveswithin the last two contention slots of the current MCI frame, in order to start serviceimmediately (by sending or scheduling a request) the following condition should besatisfied,Current _ time + D + D ≤ CSn(4.11)Sl _ TxpropTx _ timewhere Current_time is the time when the packet arrives and CSn Tx_time is the time atwhich the n th contention starts. If this condition is not satisfied, the DVB/DAVICreservation mechanism retries with the contention slots of the following MCI frame andthen selects randomly one contention slot from the set of contention slots that satisfyEquation 4.11.4-21

Chapter 4Simulation and analytical modelling pFirst we analyse the case when the service times of the data packets are either X 0 or X 1 ,and consider for the moment that packets that arrive at the end of the MCI cycle aregiven a service time of X 0 (instead of X 1 ). Once we have derived a formula for the meanservice time, we can easily obtain a similar equation, but now taking into account thethird case of service time X 2 (=X 1 ).The initial mean service time, let’s sayX ' , is given by the probability that a data packetfinds the system idle, P 0 ’, multiplied by the service time when the system is idle, X idle=X 0 , plus the probability that a data packet finds the system busy, P 1 ’, multiplied byX busy = X 1 . Thus the initial service time is given by:X' = P ' ⋅ Xidle+ P1' ⋅ X busy0(4.12)where the probability of finding the system busy or idle is given by Equations 4.13 and4.14, respectively.PλP ' 1=(4.13)µ0' P1= 1−'(4.14)The service time when the system is in idle state can be obtained by making a closeranalysis in Figure 4.13, as depicted in Figure 4.14. Here X idle consists of three differentdelays, as indicated by Equation 4.15.X idle = X t1 + X t2 + X t3 (4.15)The first delay, X t1 , is a variable delay but can be approached by using the mean interarrivaltime in the range [MCI 0_start – MCI 0_end ] given by Equation 4.16.MCIMCI 0_start MCI 0_end= MCI 2_end1_start MCI 1_end= MCI 2_startD Pk0MCI 0MCI 1 MCI 21 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9R D DX t1X t2X t3A Pk0X idleContention SlotRContention Slot withReservation RequestDData SlotFigure 4.14 – Mean service time when the system is in idle state.4-22

Chapter 4Simulation and analytical modelling pMCI0 _ end− MCI0 _ start MCItXt1= =(4.16)22where MCI t is the duration of a complete MCI frame (= 0.75 ms) that describes 9contention slots.The second delay, X t2 , is referred to as ‘scheduler look-ahead’ time. The number ofMCI frames, MCI , that the INA scheduler will have transmitted between subsequentgrants is directly related to the time that an MCI frame is scheduled before the currentMCI frame ends. This delay is given by:Xt= MCI⋅ MCI(4.17)2 tAs illustrated in Figure 4.15, the INA scheduler does not wait until the current MCIframe expires before it transmits the next one. This is to be expected since there is adelay associated with the transmission and propagation delay of the MCI frame, as wellas a processing delay for parsing the MCI frame in each NIU. Consequently, the INAmust make sure that the next MCI frame is sent soon enough in order to reach the NIUs,and that it is processed before the time described by the previous MCI frame elapses.In order to achieve this functionality the scheduler needs to take into account thefollowing possible delays that an MCI frame will experience in the worst case scenario:• Downstream transmission delay of an MPEG2 data frame (D tx_MPEG ). This delay isincurred if a data message has just started being transmitted before the MCI frame isscheduled and cannot be cancelled. For a 42 Mbps downstream channel D tx_MPEG =INANIUMPEG2-framewith MCI info.d lahMPEG2-framewith MCI info.Time describedper MCI frameFigure 4.15 – Scheduler look ahead delay.4-23

Chapter 4Simulation and analytical modelling p204*8/42M = 38.9 µs.• Transmission delay of an MPEG2 frame containing an MCI frame (D tx_MPEG ). Thisallows the scheduler to take into account transmission delay of the forthcoming MCIframe.• Interleave delay (D interleave ). This delay is incurred to enable the correction of burstnoise induced errors, protecting against a burst of symbols errors from being sent tothe RS decoder.• Roundtrip propagation delay (2·D prop ). This is the delay incurred to transmit a singlebit of information from the INA to the furthest NIU and vice-versa.• NIU processing delay (D IU_proc ). This is the delay that it takes to the NIU to parsethe MCI frame and translate it into transmission opportunities.Based on the above delays, the scheduler look-ahead delay, D lah , is given by:Dlah= 2 ⋅ D + D + ⋅ D + D(4.18)Tx _ MPEG int erleave2propIU _ procAccording to [34], the interleave delay is given by the following equation,Dint erleaveDS _ size ⋅8⋅(interleave _ depth −1)= (4.19)Dwsbit _ ratewhere DS_size (=204 bytes) is the total size of the MPEG-2 TS frame to be transmitteddownstream including FEC information and the interleave_depth is a constant set to 12for downstream IB modulation. D ws_bit_rate is the downstream data rate set to 42 Mbps.With these values indicated above, the interleave delay results in 427 µs.In order to get the complete scheduler look-ahead time (given by Equation 4.18), theprocessing time of the MCI frame at the NIU (D IU_proc ) is needed. Such delay ishardware specific, but according to the specification [34], this value should be under500 ms. We first consider a short processing delay. Assuming that this delay is about100 ms, then the scheduler look-ahead delay becomes:D lah= 2 ⋅38.9+ 427 + 2 ⋅50+ 100 ≈ 705µs4-24

Chapter 4Simulation and analytical modelling pThe number of MCI frames, MCI (needed in Equation 4.17) that would be scheduledbefore the current MCI frame expires is then calculated as D lah over the length of thesubsequent MCI frames (MCI t ). Thus:⎧ Dlah⎪when Dlahmod MCIt= 0MCItMCI= ⎨(4.20)⎪ Dlah+ 1 otherwise⎪⎩MCItFor this analysis (when D IU_proc ≈ 100 ms), the scheduler look-ahead delay results inD lah = 705 µs, hence MCI = 705·10 -6 / 750·10 -6 ≈ 1. Consequently, the secondcomponent of delay of the service time in idle state results in Xt= 1⋅MCI .2 tOn the other hand, if we consider a large processing delay (e.g D IU_proc ≈ 500 ms), thenthe second component of delay of the service time becomes:−6−6Xt2= 1105 ⋅10/ 750 ⋅10≈ 2 ⋅ MCI tIn order to get a complete formula for X idle we now need to calculate the thirdcomponent of delay, X t3 , which is related to the number of MCI frames required totransmit a data packet (Pk mci ) and is given by:X3= Pk ⋅ MCI(4.21)where Pk mci can be calculated by using the following equations:tmcitPkslot⎧ Pk⎪= ⎨⎪ Pk⎪⎩sizesize+ AAL5AAL5PDU+ AAL5AAL5PDUheaderheader+ 1when Pksize+ AAL5headermod AAL5PDU= 0otherwise(4.22)In this equation, Pk slots gives the number of upstream slots required to transmit a datapacket, Pk size is the length of the packet size in bytes to be transmitted, AAL5 header (= 8bytes) corresponds to the protocol overhead caused by the encapsulation of PDUmessages into ATM cells, AAL5 PDU (= 48 bytes) is the payload of an ATM cell.4-25

Chapter 4Simulation and analytical modelling pThen, the number of MCI frames required to transmit a data packet is Pk slots over themaximum number of reservation slots per MCI frame. Thus:⎧ Pkslot⎪when Pkslotmod MCImax_ slots− MCICSs= 0MCImax_ slots− MCICSsPk = ⎨(4.23)mci⎪ Pkslot+ 1 otherwise⎪⎩ MCImax_ slots− MCICSswhere MCI max_slots and MCI CSs correspond to the maximum number of upstream slotsdescribed per MCI frame and the minimum number of contention slots per MCI frame,respectively.By substituting X t1 , X t2 and X t3 from Equations 4.16, 4.17 and 4.21 respectively inEquation 4.15, the service time in an idle state now becomes:XidleMCIt= + MCI⋅ MCIt+ Pkmci⋅ MCIt= (0.5 + MCI+ Pkmci)2⋅ MCIt(4.24)The service time in a busy state (X busy ) is similar to X idle , but the only difference is thatnew packets, finding the system busy, spend a complete MCI frame looking for acontention slot in which to place a reservation request instead of MCI t /2. Thus, thecalculation for X busy is given by:Xbusy= MCI + X2+ X = (1 + + Pk ) ⋅ MCI(4.25)ttt3MCImcitUp to this point, we have calculated all the elements of the initial mean service time.Thus by substituting Equations 4.13, 4.14, 4.24 and 4.25 in Equation 4.12,calculated as:X ' is thenHence,λλX ' = (1 − ) ⋅ (0.5 + MCI+ Pkmci) ⋅ MCIt+ ⋅ (1 + MCI+ Pkmci) ⋅ MCIµµt⎛λ ⎞X ' = ⎜ Pkmci+ 0.5 + MCI+ ⎟ ⋅ MCI(4.26)t⎝2 ⋅ µ ⎠We can now derive the real mean service time ( X ) when the third case of service time(X 2 ) is considered. Since not all newcomer packets finding the system in idle state are4-26

Chapter 4Simulation and analytical modelling pgiven a service time of X idle , the probability P 0 that a service time of X idle is actuallygiven, needs to calculated. This probability is obtained by the following equation:P0( MCIt− DSl_ Tx− Dprop)= P0'(4.27)MCItThus, the probability P 1 that a data packet is given a service time of X busy is thenP1 = 1−P 0(4.28)Therefore, the real mean service time results in:X= P X0idle+ P X1busy= (1 −+λµ) ⋅ (MCIt−DSl_ Tx −DpropMCItλ MCIt−DSl_ Tx −Dprop( 1- (1 − ) ⋅ ()) ⋅ (1 + MCI+ Pkmci) ⋅ MCItµMCIt) ⋅ (0.5 + MCI+ Pkmci) ⋅ MCIt+Hence,Xλ= ( DSl_ Tx+ Dprop) ⋅ (0.5 −2⋅ µ) + (0.5 + MCI+ Pkmci+2⋅µ)λ⋅ MCI (4.29)tOnce we have calculated the mean service time ( X ), we can now obtain the variance2( σ ), needed in Equation 4.10. It is well-defined that the variance is the averageXsquared deviation from the mean, given by the following formula:∑( X− X )2jj== 12σ (4.30)XIf we consider that all packets receive a service time either of X idle or X busy , then thevariance can be re-calculated as:= ⋅ ( X − X ) + (1 − ρ)⋅ ( X − X )222σ ρXbusyidle(4.31)In this new equation, the variance only depends on the value of the utilisation factorλ( ρ = µ< 1). Thus, the variance follows the distribution depicted in Figure 4.16.4-27

Chapter 4Simulation and analytical modelling pFrom this figure we can appreciate that the highest value for the variance is when ρ =0.5, which results in σ 2= 0.000000036 ≈ 0 . By substituting σ 2 = 0 in Equation 4.10,Xthe mean waiting time in queue is then given by:λ ⋅ X 2W =(4.32)2 ⋅ (1 − ρ)By substituting Equations 4.29 and 4.32 in Equation 4.4, the derivation of the end-toendpacket delay is complete and is given by:λλλ ⋅ XD ete = ( DSl_ Tx+ Dprop) ⋅ (1.5 −2⋅µ) + (0.5 + MCI+ Pkmci+2⋅µ) ⋅ MCIt+ (4.33)λ2 ⋅ (1 − )X2µ4.4.2 Throughput formulation for a single node scenarioHaving formulated the total end-to-end packet access delay, the throughput, S, is givenby the number of packets serviced per second (L x ) multiplied by the packet size, thus:S = L x⋅ Pk size(4.34)In order to obtain L x , we need to calculate first the mean number of packet serviced perbusy cycle,L bc . Then, the number of packets serviced per second can be easilyobtained by dividingL bc between the average time of a busy cycle, T bc, thus:LbcLx= (4.35)T bcFrom analysing Figure 4.17 a busy cycle ( Tbc) can be defined as the sum of a busyLσ 2 X (x10 -9 )a rriv a ld e p a rtu re5 04 03 02 01 0T bp 1 T ip1 0 T b p2T ip 2T bp 3TT bp n ip30 0 .1 0 .2 0 .3 0 .4 0 .5 0 .6 0 .7 0 .8 0 .9 1T b c 1T b c 2T b c 3T b c nρ = λ µFigure 4.17 – Time of busy cycle.Figure 4.16 – Variance of the mean service time.T ip ntim e4-28

Chapter 4Simulation and analytical modelling pperiod, T bp , plus an adjacent idle period, T ip , this gives:T +bc = T bp T ip(4.36)In Equation 4.36, the busy period ( Tbp) is defined to begin with the arrival of a datapacket to an idle channel and ends when the channel next becomes idle. Accordance toGross [23], this busy period can be found by using the ratio:TTbpipPr obability _ that _ the _ system _ is _ busy==Pr obability _ that _ the _ system _ is _ idle 1−λµλµSince the arrivals are assumed to follow a Poisson distribution, the idle period isexponential with mean1T ip= . Thus, the busy period is given by:λλµ11 1T bp =λ⋅ = =(4.37)λλ1−⋅ (1 − ) µ − λµµµBy substituting Equation 4.37 in Equation 4.36, the busy cycle results in:1 1 µ 1T bc = T bp + T ip = + ==(4.38)λµ − λ λ λ ⋅ ( µ − λ ) λ ⋅ (1 − )The number of packets served in a busy cycle is now considered. This value can beobtained by making the following analysis. If a data packet is served in X units, thenthe number of packets served in the busy cycle is given by dividing the average time ofthe busy period between the average time that it takes a packet to be served, so:1Tbpµ − λ 1 1Lbc = = ==X X X ⋅ ( µ − λ)1−(4.39)Finally, the number of packets served per second is then calculated as:1λL bc(1 −µ)Lx= =T bc1= λ(4.40)λ(1 − )λµTherefore our formulation now is complete and the throughput can be obtained by:λµµ4-29

Chapter 4Simulation and analytical modelling pS = λ ⋅ Pk , ∀ρ< 1(4.41)size4.4.3 Bounds on maximum system throughputSo far, we have formulated the mean access delay and the throughput for a single nodeconfiguration. However, for a larger number of nodes we have derived some boundsthat can be used to estimate the maximum system throughput sustainable by theDVB/DAVIC’s upstream channel. These bounds depend mainly on the traffic typesbeing delivered and on the maximum bandwidth assigned for contention and reservationaccess, as well as the contention resolution algorithm used. In this section we presentbounds for three different configurations. These bounds will play a vital role to validatemore complex traffic configurations in the following chapters.A simple bound on the maximum upstream throughput would be when only data slotsare being transmitted, so that 48 bytes out of 64 bytes are transmitting payload for anefficiency of (48/64=) 75%. This would be approached in theory if very large packetswere transmitted, so that the capacity of the contention-based access region wasnegligible. But in practice, part of the bandwidth is reserved for contention access, andthe efficiency heavily depends on the bandwidth assigned to this access mode.Therefore, the maximum system efficiency can be estimated by Equation 4.42.Smax_ l arg e _ packets48⋅RS= (4.42)64 ⋅ ( RSs + CSc)where RSs and CSs are the number of reservation and contention access slots describedper signalling frame, respectively.The simple bound of the maximum upstream throughput presented in Equation 4.42cannot be used to calculate the maximum throughput of variable packet sizes. This isbecause such estimation was based only on large packets being transmitted when mostof the bandwidth was allocated to the reservation-based access region and did not takeinto account the additional bandwidth that should be allocated to the contention-basedaccess region so as to resolve collision. In [96] and [98] a more efficient calculation toestimate the maximum throughput of a CATV network (based on the IEEE 802.144-30

Chapter 4Simulation and analytical modelling pstandard) was proposed. By making slight changes we can use the same estimationapplied to the DVB/DAVIC standard as shown in Equation 4.43.PayloadPksizeSmax_ exp_ backoff==(4.43)Payload+Overhead+eContention _ Slots Pk + eCSsFor the splitting tree algorithm, reservation requests are now transmitted using aminislot (MSs) of 21 bytes instead of a complete contention slot of 64 bytes. Thus, themaximum theoretical system throughput that can be achieved with this algorithm isgiven by Equation 4.44.PksizeSmax_ split _ tree=(4.44)Pk + eMSsslotsslots4.5 Access delay and system throughput validationIn this section we validate the results yielded by the simulation model for differentpacket sizes and variable offered load. In order to get the maximum throughput, we setthe ‘Minimum number of contention slots per MCI frame’ to 1 slot. The arrival rate ofthe packets was exponentially distributed and the data rate for the upstream channel wasat 6.17 Mbps. Table 4.2 presents the maximum throughput sustainable for differentpacket sizes ranging from 64 to 1518 bytes.In this table the maximum throughput is given by Equation 4.45.SmaxPk ⋅8= (4.45)Xwhere X corresponds to the maximum service time given by Equation 4.29.4-31

Chapter 4Simulation and analytical modelling pFigures 4.18 and 4.19 show the average end-to-end delays and throughput forsimulation and analytical results, respectively, as the offered load in increased for thesix different packet sizes presented in Table 4.2. For each packet, the offered load (λ)was increased from a relatively low rate at 16 kbps, up to approximately 98% of themaximum theoretical throughput. Offered loads close to 100% of the maximumthroughput were not considered, since the stability of the M/G/1 systems [10] requiresλ < 1.Table 4.2 – Maximum theoretical throughput for different packet sizes.Packet Size (bytes)Throughput (Kbps)3000270024002100180015001200900600300Max. Service Time X (ms)Max. Throughput (S max )64 2.24 228 kbps (≈ 4 % of the cc)128 2.24 456 kbps (≈ 7 % of the cc)256 2.24 912 kbps (≈ 15 % of the cc)512 2.99 1368 kbps (≈ 22 % of the cc)1024 3.74 21876 kbps (≈ 35 % of the cc)1518 Throughput 4.49 for Simulation and Theoretical 2703 kbps (≈ Results 44 % of the cc)0Simulation ResultsTheoretical Results64-bytepacket256-bytepacket128-bytepacket1024-bytepacket512-bytepacket1518-bytepacket0 250 500 750 1000 1250 1500 1750 2000 2250 2500 2750 3000Offered Load (Kbps)Figure 4.18 – System throughput: simulation and theoretical resultsfor different packet sizes.4-32

Chapter 4Simulation and analytical modelling pBy analysing these figures, we can verify that the results obtained using the simulationmodel for the DVB/DAVIC protocol are in good agreement with theoretical results. Forexample, for the worst case scenario that is when a packet size of 64 bytes is used. Fromsimulation results (Figure 4.18), the maximum throughput that a single node canachieve resulted in 224 kbps. From analytical results (Table 4.2), this number resulted in228 kbps. This extremely low performance of the DVB/DAVIC protocol (equivalent to≈ 4% of the channel capacity) is due to the two-phase data transmission cycle, whichrequires a request to be issued in the upstream channel and the data to be transmittedafter a grant message is received in the downstream channel.The delay between the time when a request is issued and the time when it is granted isanother parameter, and needs to be analysed and accurately calculated. This delay,referred here as the ‘scheduler-look ahead (D lah )’, has proven to be one of the majordelay elements in the transmission cycle, which dramatically reduces the throughput.When the packet size was changed to maximum size (1518 bytes) the maximumthroughput sustainable per station resulted in ≈ 2630 kbps (equivalent to 43% of thechannel capacity).In general, results for mean access delays (Figure 4.19) revealed that for all packetsanalysed, the maximum deviation between simulation and analytical results was foundMean Access Delay (ms)100908070605040302010064-bytepacketMean Access Delays for Simulation and Theoretical ResultsSimulation ResultsTheoretical Results128-bytepacket256-bytepacket512-bytepacket1024-bytepacket1518-bytepacket0 250 500 750 1000 1250 1500 1750 2000 2250 2500 2750 3000Offered Load (Kbps)Figure 4.19 – Mean access delay: simulation and theoretical resultsfor different packet sizes.4-33

Chapter 4Simulation and analytical modelling pMean Access Delay (ms)Mean Access Delay (ms)807060504030201001009080706050403020100(a) Meand Access Delay for 64-byte packetsTheoreticalSeed-44Seed-131Seed-224Seed-257Seed-300Seed-452Seed-569Seed-664Seed-742Seed-809Seed-9480 25 50 75 100 125 150 175 200 225 250Offered Load (kbps)(c) Meand Access Delay for 128-byte packetsTheoreticalSeed-44Seed-131Seed-224Seed-257Seed-300Seed-452Seed-569Seed-664Seed-742Seed-809Seed-9480 50 100 150 200 250 300 350 400 450 500Offered Load (kbps)Throughput (kbps)Throughput (kbps)2502252001751501251007550250500450400350300250200150100500(b) Throughput for 64-byte packetsTheoreticalSeed-44Seed-131Seed-224Seed-257Seed-300Seed-452Seed-569Seed-664Seed-742Seed-809Seed-9480 25 50 75 100 125 150 175 200 225 250Offered Load (kbps)(d) Throughput for 128-byte packetsTheoreticalSeed-44Seed-131Seed-224Seed-257Seed-300Seed-452Seed-569Seed-664Seed-742Seed-809Seed-9480 50 100 150 200 250 300 350 400 450 500Offered Load (kbps)Figure 4.20 – Mean access delay and throughput for different seeds.to be under 3% with an offered load up to ≈ 98% of the maximum channel capacity.This deviation was found to be under ≈ 2% for system throughput (Figure 4.18).Simulation results presented in the previous two figures were obtained with a seed of257.We also used different seeds for simulation results in order to verify the accuracy ofthe simulation model. Figure 4.20 shows the performance analysis when ten randomseeds were used (selected in the range from 0 to 1000) for 6 4 and 128 byes packetsizes. Results for mean access delays revealed that all seeds follow the same distributionas that of the theoretical results. For offered loads under 95% of the maximumtheoretical throughput, the maximum deviation was still under 3%. For offered loadsclose to saturation (from 96 to 99%) the deviation was seen to be about 4%. However,the maximum deviation in terms of throughput was unchanged at 2%.4-34

Chapter 4Simulation and analytical modelling pWe have proved that simulated performance of the system does not depend on the seedused, therefore it will not be mentioned again and all simulation results presented in thefollowing chapters will be yielded using a seed of 257.For a larger number of nodes, we also carried out a rigorous verification test of theoperation of the DVB/DAVIC simulation model. For this test, we used a mixed trafficpattern consisting of 32 kbps IP traffic and 9.7 kbps VoIP traffic (these traffic types aredescribed in Section 5.4), using the exponential backoff algorithm. We ran a simulationfor 60 seconds and captured the sequence of events between the 59 th and the 60 thsecond. Appendix D shows this sequence for 30 stations. From analysing these eventswe can verify that every station transmitted all packets received and used the contentionresolution algorithm correctly. On average, every station received (1/(0.092 IP interarrival+ 0.12 VoIP inter-arrival) = 19.6 packets per seconds, which corresponds to thenumber of packets received and transmitted in Appendix D.4.6 ConclusionsThe DVB/DAVIC protocol is a relatively new specification and little information aboutits performance and scalability is known. In this chapter, the suitable simulation andtheoretical models have been presented. These models will be used in subsequentchapters for the performance evaluation, optimisations and enhancements of theDVB/DAVIC protocol.The simulation model implemented for this research was based on the main networktopology of the CSF (v.13), developed by MIL3, Cablelabs and the University ofSheffield. This model has been significantly modified to support the basic functionalityof the MAC and PHY layers of the DVB/DVIC protocol, including the functionality ofthe two contention resolution algorithms adopted by the DVB/DAVIC protocol. Inaddition, new features for further optimisations were also incorporated in the simulationmodel, such as contention slot allocators, enhanced reservation requests mechanisms,and a prioritised scheduler, which will be used in subsequent chapters.Existing analytical models from the literature were insufficient to model theDVB/DAVIC protocol, as they did not take into account the variable bandwidth of the4-35

Chapter 4Simulation and analytical modelling pcontention-based access region and the performance of the exponential backoffalgorithm or the splitting tree algorithm. A less complicated theoretical model has beenformulated for the purposes of validation of the simulation model. Such a model wasbased on an M/G/1 queuing system for a single node scenario. This model addressedone of the fundamental performance properties of computer communications protocols,which relates to the mean packet access delays and maximum sustainable throughput ofa station.Results obtainable using the simulation model for the DVB/DAVIC protocol were ingood agreement with theoretical results, with a maximum deviation of 4% for meanaccess delays and 2% for throughput. We also found that the DVB/DAVIC protocol ishighly inefficient when short data packets are transmitted, achieving only ≈ 224 kbps(≈ 4% of the channel capacity) when 64-byte packets are delivered.For a larger number of nodes, we have also provided some theoretical bounds that couldbe used to estimate the maximum system throughput sustainable by the DVB/DAVIC’supstream channel for different traffic loads and protocol configurations.4-36

Chapter 5UPSTREAM CHANNEL CAPACITYAND CHARACTERISATION5.1 IntroductionThe design and implementation of communications protocols require knowledge offundamental performance characteristics. Such characteristics include channel capacity,system throughput and utilisation bounds, maximum number of nodes (or streams)supported, mean access packet delay and the maximum throughput achievable per nodeunder different traffic loads and configurations, among others. Knowledge of thesecharacteristics allows service providers or network operators to plan the offeredservices, make cost and tariffing predictions and implement the network topologies. Inaddition, there are protocol operational parameters for which only recommended valuesare given in the specification. Such parameters should be optimised for different trafficconditions in order to achieve optimum system performance.From the literature survey presented in Section 1.2.3.3, it may be postulated that there isa lack of studies that address the scalability, performance issues and optimisation of theDVB/DAVIC protocol explicitly. Therefore, the objective of this chapter is to present acomprehensive estimation of the performance characteristics of the upstream channel,when the contention and reservation access modes are used. In order to study complexanalysis and address issues of capacity and scalability for the DVB/DAVIC protocol,the simulation model described in Section 4.3 will be used. The structure of the analysisto follow is presented in the following section.5-1

Chapter 5Upstream channel capacity and characterisation p5.2 Case studiesHaving introduced and validated the OPNET simulation model in Chapter 4, a numberof analyses are presented here to define the fundamental properties and scalability of theupstream channel. Specific issues of the case studies address the following performanceanalysis given below.• Offered load scalability: how the protocol scales against increasing the offeredload generated by a number of active stations, and what the delay and systemthroughput characteristics are.• Capacity in terms of active stations: how the protocol scales against anincreasing number of active stations, each generating a predefined traffic load.• Station’s buffer capacity: how the performance characteristics change if thestations’ buffer capacity is altered.• Effects of contention slots for data transmission: what the delay and systemthroughput characteristics are when transmitting a determined number of cells inthe ‘contention-based access region’.• Effects of varying the maximum reservation request message per station: howthe protocol scales against an increase in the maximum ‘reservation requestmessage’.• Effects of changing the number of MCI frames: performance gain when thesignalling frame cycle is reduced.• Effects of varying the packet size in isochronous streams: how the protocolscales against changing the packet size and inter-arrival times when deliveringisochronous data rates.These configurations are examined because they demonstrate the type of results thatcould be obtained using both the simulation and analytical model, and also help toidentify key issues and suggest performance optimisation methods with a wide range ofapplications. Results, such as maximum network capacity, can be used for capacity5-2

Chapter 5Upstream channel capacity and characterisation pplanning as well as identifying the limits of the different levels of service that can beprovided to subscribers.For all the configurations defined above we need to find saturation points and thereasons for inefficiencies, as well as how optimal features such as the use of the splittingtree algorithm to resolve collisions could possibly improve the overall systemperformance.5.3 Traffic type characteristics and system parametersAs stated in Section 4.3.2, each station has been enabled with two traffic sources, thefirst traffic source generates Internet traffic and the second generates either VoIPstreams or isochronous streams. In this chapter we base our analysis on these two trafficsources to examine the impact and trade-offs on system performance of four differenttopical traffic types: Internet traffic, Voice over IP (VoIP), mixed traffic, andisochronous streams, (described below) and at the same time providing a performancecharacterisation of the DVB/DAVIC protocol.5.3.1 Internet trafficThis generator emulates Internet traffic. From available traffic measurement studies [46]and [53], it is known that the frequency of Ethernet packets is as indicated in Table 5.1,which also presents the number of ATM cells required per packet and the total packetsize to be transmitted. It is expected that most upstream packets will beacknowledgements of the higher bandwidth downstream packets. Hence, the majority ofpackets generated will be 64 bytes long and packet sizes of 1518 bytes correspond to themaximum size of Ethernet packets that are mainly used by File Transfer Protocol (FTP)applications.Packet Size(bytes)Table 5.1- Packet size distribution and characterisation.Probability ofoccurrenceumber ofATM cellsTransmitted packetsize with overhead64 0.6 2 128 1128 0.06 3 192 1256 0.04 6 384 1512 0.02 11 704 11024 0.25 22 1408 25-3umber of signallingframes required

Chapter 5Upstream channel capacity and characterisation p1518 0.03 32 2048 2Packets are generated with exponential distributed inter-arrival times, with a mean valueselected to produce the desired average data rate as defined in [53]. In this chapter theinter-arrival times are set in such a way that the resulting mean offered load per activestation is 64 kbps, 32 kbps, 26.6 kbps or 16 kbps according to the configurationanalysed. The average packet size (Pk _bytes ) for this traffic type is ≈ 368 bytes, whichcomputes to 8.3 ATM cells (Pk _slots ) per message.5.3.2 Voice over IP (VoIP)Today voice is transmitted either in analogue or digital form in circuit-switchednetworks (PSTN, ISDN and GSM). Between two end users a reserved (virtual) channelexists. A constant data rate is used to transmit the voice data (e.g. 64 kbps in theEuropean ISDN, 56 kbps in the USA ISDN) with a service charge. In future, voice callsmay be transmitted via existing IP networks with the same or better levels of quality.We have selected VoIP traffic (from the set of services of ‘IP Telephony’) to analyse theperformance characterisation of the upstream channel, because this technology will playa key role in future telecommunications networks. In general, IP Telephony refers tocommunications services (e.g. VoIP, fax and voice-messaging applications) that aretransported via the Internet in digital form using discrete packets rather than in thetraditional circuit-committed protocols of the Public Switched Telephone Network(PSTN). The VoIP traffic type, presented here, emulates a speech codec ‘G.723.1’ [57],which according to the ITU, IETF and the VoIP Forum is the preferred codec forInternet telephony applications [101] and [28].This codec generates a data rate of 5.3 kbps or 6.3 kbps depending on the mode. In thisresearch codecs of 5.3 kbps will be used. This codec generates and encodes a 20-bytedata frame every 30 ms. In a study carried out in [28], it was found that as the numberof audio frames per packet increases, the packet overhead decreases. The overheaddecrease is explained by more information bytes (audio frames) being included in apacket with a fixed header (RTP+UDP+IP). Conversely, the latency increases as thenumber of frames per packet increases. This is expected since more audio frames5-4

Chapter 5Upstream channel capacity and characterisation prequire more time to be captured and buffered. Hence, a trade-off exists between packetoverhead and local latency for audio packet transfer.The optimal point for the number of audio frames per packet is between seven and eightfor G.723.1 codecs. According to [28], these values should be used by terminals thattarget packet efficiency as well as low latency for audio packets. However, sometimes,the audio local latency is reduced at the expense of packet overhead, in order to achievelatencies acceptable to users. In fact, a few applications are willing to sacrifice someprotocol overhead for achieving better audio latency. Some H.323 terminals use a valueof 3 or 4 for the number of frames per audio packet in order to reduce this latency [28].Therefore, in order to be consistent with these figures for H.323 terminals, 4 audioframes will be used per packet. Then, the following protocol overheads are added inorder to yield a complete VoIP packet.Each audio packet has a Real Time Protocol (RTP) header of 12 bytes that carriessequence numbers, a synchronisation source identifier and time-stamps. A UserDatagram Protocol (UDP) header of 8 bytes is needed to carry a UDP with an unreliabletransmission service. In addition a 20-byte IP header is needed to transfer routinginformation. A 3-byte Logical Link Control (LLC) and a 5-byte SubNetworkAttachment Point (SNAP) headers are used to carry a PDU with a connectionlessservice and to identify the type of the bridged media, respectively. Finally, an 18-byteMAC header is used to transmit the PDU to its final destination over the shared media.Thus, by adding the complete headers one obtains an improved VoIP stream of 9.7 kbpsas indicated in Table 5.2. Hence, in our performance analysis, VoIP streams of 9.7 kbpswill be considered.Table 5.2- VoIP encapsulation with and without header suppression.Frame/Header 9.7 kbps Streams 12.4 kbps StreamsVoice frame 80 bytes 20 bytesRTP headerUDP header6 bytes12 bytesIP header 8 bytes 20 bytesLLC headerSNAP header3 bytes5 bytesEthernet MAC header18 bytes5-5Total Size 146 bytes (4 ATM cells) 40 bytes (1 ATM cells)

Chapter 5Upstream channel capacity and characterisation pAnother novel and topical VoIP stream that will be analysed is that of 12.4 kbps, whereevery 30 ms a voice frame is generated and encoded using header compression, asillustrated also in Table 5.2.In order to support VoIP traffic, according to [118], there is a need for an overall end-toendpacket delay of less than 150 ms for a high-quality call and up to 400 ms for a lowservicequality call. In our experiments we consider delays under 50 ms from the NIU tothe headend for the support of VoIP streams, leaving an extra 100 or 350 ms delay forthe final destination according to the expected quality of the call.5.3.3 Mixed traffic (Internet +VoIP)This traffic type emulates a combined traffic situation, where each station is generatinga VoIP data stream of 9.7 kbps, as introduced in Section 5.3.2. Additionally, someInternet traffic, as presented in Section 5.3.1, is multiplexed into the data stream and istransmitted via the upstream channel. The mean data rate per active station is set to32 kbps (consisting of 9.7 kbps of VoIP traffic and 22.3 kbps of Internet traffic) or 41.7kbps (consisting of 9.7 kbps of VoIP traffic and 32 kbps of Internet traffic) according tothe case study analysed.5.3.4 Isochronous streamsIsochronous streams are time-dependent and exist with processes where data must bedelivered within certain time constraints. For example, most multimedia streams requirean isochronous transport mechanism to ensure that data is delivered as fast as it isdisplayed and that the audio is synchronised with the video. This traffic type emulatesisochronous streams with data rates of 12 kbps, 32 kbps, 64 kbps and 128 kbps, suitablefor timing-critical interactive services (e.g. low quality video,compressed/uncompressed voice telephony and audio). Different packet sizes (64, 128,256, 512, 1024 and 1518 bytes) were considered for the delivery.All isochronous streams used in the simulations included the higher layer protocoloverhead. For example, the most likely protocol for isochronous streams is TCP/IP witha Direct IP or Ethernet bridge solution. The latter has (at the MAC layer) a 61-byteoverhead comprising of 20-bytes (TCP header) + 20-bytes (IP header) + 3-bytes (LLC5-6

Chapter 5Upstream channel capacity and characterisation pheader) and 18-bytes (Ethernet MAC header/trailer). Therefore, not all of the streamcapacity is available to isochronous applications. For instance, in the worse casescenario when 64-byte packets are delivered, for a 12 kbps isochronous stream theeffective bandwidth (excluding higher layer protocol overhead = 20-bytes TCP + 20-bytes IP + 3-bytes LLC + 18-bytes Ethernet MAC = 61-bytes), is dramatically reducedto 0.56 kbps when the TCP/IP protocol with an Ethernet bridge solution is considered.Such overheads should be taken into account when considering the delivery ofapplications with specific bandwidth requirements. The effective throughput for thedifferent streams as a function of the packet size and protocol stack (see Figure B.1)used for these simulations is given in Table 5.3, (shaded lines in the table present theeffective bandwidth when the TCP/IP with an Ethernet bridge solution is considered,which is the most popular solution in CATV networks).It is evident from Table 5.3, that although lower packet sizes provide quickerinteraction, smaller packet sizes reduce the available bandwidth due to a proportionalincrease in overhead. In order to compensate for the considerable reduction inbandwidth, in the last version of the of the DVB/DAVIC protocol [35] (recentlyTable 5.3 - Effective bandwidth in isochronous streams (kbps).Isochronous Streams (kbps)Packet size (bytes)64 128 256 512 1024 151812 TCP/IP 3.94 7.97 9.98 10.99 11.50 11.66UDP/IP 6.19 9.09 10.55 11.27 11.64 11.75PPP 9.38 10.69 11.34 11.67 11.84 11.89TCP/MAC Eth 0.56 6.28 9.14 10.57 11.29 11.29UDP/MAC Eth 2.81 7.41 9.70 10.85 11.43 11.6132 TCP/IP 10.50 21.25 26.63 29.31 30.66 31.09UDP/IP 16.50 24.25 28.13 30.06 31.03 31.35PPP 25.00 28.50 30.25 31.13 31.56 31.70TCP/MAC Eth 1.50 16.75 24.38 28.19 30.09 30.71UDP/MAC Eth 7.50 19.75 25.88 28.94 30.47 7.5064 TCP/IP 21.00 42.50 53.25 58.63 61.31 62.19UDP/IP 33.00 48.50 56.25 60.13 62.06 62.69PPP 50.00 57.00 60.50 62.25 63.13 63.41TCP/MAC Eth 3.00 33.50 48.75 56.38 60.19 61.43UDP/MAC Eth 15.00 39.50 51.75 57.88 60.94 15.00128 TCP/IP 42.00 85.00 106.50 117.25 122.63 124.37UDP/IP 66.00 97.00 112.50 120.25 124.13 125.39PPP 100.00 114.00 121.00 124.50 126.25 126.82TCP/MAC Eth 6.00 67.00 97.50 112.75 120.38 122.86UDP/MAC Eth 30.00 79.00 103.50 115.75 121.88 30.005-7

Chapter 5Upstream channel capacity and characterisation preleased in October 2001), a new advanced mechanisms is used for header suppression.By using this feature, all constant components of the different headers can besuppressed, thus the effective data rate of isochronous streams (e.g. multiplexed voice,audio and/or video) is remarkably increased.5.3.5 System parameters and assumptionsIn all simulations, one upstream channel with a capacity of 3.088 Mbps and onedownstream channel with 42 Mbps were considered. Another important simulationparameter is the choice of the network size. According to [100] the number of activesubscribers per upstream channel on European cable networks is approximately 50 usersin peak hours of traffic.In our analysis we use cluster sizes of up to 70 stations per upstream channel in order toanalyse the performance of the DVB/DAVIC protocol during high periods ofcongestion. Some scenarios will approach an increased number of stations (up to 340stations), with a reduced traffic load to analyse the maximum number of nodes that canbe supported. The complete set of simulation parameters used in order to produce theresults presented in this chapter are summarised in Table 5.4.Table 5.4 – Simulation parameters.Simulation ParameterValueUpstream data rate (QPSK)3.088 MbpsDownstream rate (64-QAM)42 MbpsDownstream signallingIn BandMaximum number of active NIUs (EuroModems) 340*Minimum and maximum backoff values for the exponential backoff algorithm 3 and 5*Entry spreading factor of the splitting tree algorithm 6*Minimum contention-based slots per signalling frame3 slots*Buffer capacity per NIU3000 ATM cells*Transmission time (cycle) of the signalling frame3 ms*Simulation time for each run60sMaximum reservation request message length32 ATM cells (slots)*Maximum contention access message length0 ATM cells (slots)*Distance from nearest/farthest NIU to the Headend10-16 KmSpacing between closest and farthest NIURandomly distributedHeadend and NIU processing delay2 microseconds eachPropagation delay (coax and fibre)5 microseconds/Km* These parameters will be ranged as indicated by the traffic scenarios5-8

Chapter 5Upstream channel capacity and characterisation pThe values selected for the different parameters are the defaults defined in thespecification where applicable. All references to packet sizes in the analysis that followsrefer to the size of the packet as it enters the system from a PC station. This would bethe packet that the NIU is going to submit for delivery over the cable network and doesnot include DVB MAC and PHY overheads. However, those include 18-byte Ethernet,20-byte IP, 20-byte TCP, 8-byte UDP and 12-byte RTP (for VoIP only) headers. Theseprotocol overheads should be taken into account when the results presented here areused to evaluate application performance. In addition, simulations will be run using bothcontention resolution algorithms adopted by the DVB/DAVIC protocol (exponentialbackoff algorithm and splitting tree algorithm) as defined by the scenarios.Finally, for the tests defined, the INA utilised a simple First In First Out (FIFO)scheduler. A more complex scheduler was not necessary since all the streams are treatedevenly.5.4 Performance characterisation of the DVB/DAVICprotocol5.4.1 Offered load scalabilityIn this section we analyse the system performance under increased offered load. Thesimulated network examined in this scenario consists of a small network size of 20active stations, each generating a variable amount of traffic load made up frommaximum Ethernet packets of 1518 bytes. The aim of this analysis is to find the highestachievable system performance in terms of system throughput and mean access delaysrather than the effect of packet size, which is addressed in Section 5.7.As mentioned in Section 4.4.3, a bound for the maximum upstream throughput is givenby Equation 4.42. In this analysis, we have reserved at least 2 slots (out of 18 slots) persignalling frame for contention access, CSs = 2 and RSs = 16. Therefore, the maximumtheoretical efficiency achieved is ≈ 67 %, which we may expect from our simulationanalysis (as shown below).Smax48⋅RSs==64 ⋅ ( RSs + CSs)48⋅16≈ 66.7%64 ⋅ (16 + 2)5-9

Chapter 5Upstream channel capacity and characterisation pThe network configuration for this analysis was as follows: the upstream channelcapacity (cc) was 3.088 Mbps and the offered load was ranged from a relatively low243 kbps up to 4.1 Mbps with increments of 243 kbps for each simulation run. Thecontention resolution algorithm (used to resolve collisions) was the exponential backoffalgorithm.Figure 5.1 presents the mean access delay, throughput and utilisation versus offeredload. The maximum system throughput of approximately 65% was obtained with anoffered load equivalent to the full cc (3 Mbps), which is quite close to the maximumtheoretical throughput estimated (of 67%). The deviation of 2% was accounted for byadditional slots added to the contention access region. In average, 2.3 slots wereallocated for contention access due to collisions and the inability of the allocationalgorithm (at the headend) to schedule all the data slots.The maximum channel utilisation achieved was ≈ 91%. The difference betweenmaximum throughput and utilisation of 26% is attributed to the ATM protocoloverhead, DVB Physical layer protocol overhead, reservation request transmissions,collisions and retransmissions. The remaining bandwidth of 9% of the cc was wasteddue to unused contention slots. In the following scenarios and particularly is Section5.4.7, we are going to analyse this difference in a broader context.Mean Access Delay (ms)50 045 040 035 030 025 020 015 010 05 003 M b ps Up s t reamEt h ernet frame s (1 518-by te s)Backo ff Alg orit hmMean ac c es s dela ySatur ation point at62% c han nel c apac ity8 1 5 2 3 3 1 3 9 4 6 5 4 6 2 7 0 7 7 8 5 9 3 1 0 1 1 0 8 1 1 6 1 2 4 1 3 2Offe re d L oad (%)Figure 5.1 - Mean access delay, system throughput and utilisation vs. offered load.5-10Utilis ationSy tem throughp ut1 0 09 08 07 06 05 04 03 02 01 00Channel Capacity (%)

Chapter 5Upstream channel capacity and characterisation pThe mean packet delay curve is stable (producing considerably low mean access delaysunder ≈ 33 ms) up to the point of saturation, which was found to be at approximately61% of the cc, lower than the capacity achieved by the system throughput. From thispoint, even a slight increase in offered load may result in system instability, as we canobserve with an offered load approximately of 62% of the cc.Figure 5.2 gives a better insight into the delay characteristics of the system by plottingthe number of frames that experienced delays of a certain value. In this figure at 61% ofthe cc, 75% of the Ethernet packets were transmitted in less than 33 ms and the other25% under 67 ms. At 62% traffic load, the saturation point is evident, and access delayshave been considerably increased, since only 11% of the packets are now transmittedunder 67 ms. Hence, the system cannot maintain maximum throughput whilst providingbounded delay characteristics. Therefore, the maximum system throughput sustained is1.88 Mbps (61% cc). In the next scenario we address the number of stations that can besupported.1.0Load 54%0.9Cumulatitve Probability0.80.70.60.50.40.3Load 61%3 Mbps UpstreamEthernet Frames (1518-bytes)Ba ckoff AlgorithmL oad 62%Load 6 3%0.20.1Load 70%Load 85%0.00 33 67 100 133 167 200 233 267 300 333 367 400 433 467 500Me an Acce ss De lay (ms)Figure 5.2 - Cumulative probability vs. mean access delay.5-11

Chapter 5Upstream channel capacity and characterisation p5.4.2 Capacity in terms of active stationsHere, an analysis of the scalability of the cable network in terms of number of activestations is presented. For this analysis we have now considered a low trafficconfiguration per station to find out the maximum number of stations that could besupported. The analysis starts with a small network size of 20 stations, then we increasethe number of stations (in steps of 20) until the maximum network capacity has beenexceeded.The traffic load generated by each station was a single-Ethernet packet of 1518 byteswith an exponential distributed inter-arrival rate of 1 pk/sec or 12.14 kbps. Thesimulations were performed for each contention resolution algorithm.Figure 5.3 depicts a linear system throughput until the knee with respect to the numberof active stations. We can appreciate a slight increase in system performance whenusing the splitting tree algorithm. The maximum system throughput achieved was ≈61% and ≈ 65% of the cc for the exponential backoff algorithm and the splitting treealgorithm, respectively.Channel Capacity (%)10090807060504030201003 Mbps UpstreamEthernet frames (1518-bytes)12.14 Kbps kbps per stationBackoff & Tree AlgorithmThroughput-Backoff AlgorithmThroughput-Tree AlgorithmUtilisation - Backoff AlgorithmUtilisation - Tree Algorithm20 40 60 80 100 120 140 160 180 200 220 240 260 280 300 320 340umber of Active StationsFigure 5.3 - System throughput and utilisation vs. No. of active stations, exponentialbackoff and splitting tree algorithm.5-12

Chapter 5Upstream channel capacity and characterisation pIn terms of utilisation, the exponential backoff algorithm utilised approximately 86% ofthe link capacity in comparison with ≈ 90% utilised by the splitting tree algorithm. Inthis analysis, which considered the maximum Ethernet packet size, the increase insystem performance (throughput and utilisation) resulted in approximately of 4% of thecc when the splitting tree algorithm is selected. A bigger increase in systemperformance can be obtained when short packet data transmissions are used. In thefollowing analysis we change the traffic pattern to appreciate this difference better.Results for access delays, Figure 5.4 indicates that saturation points are presentedbeyond 140 stations (with mean packet delay of 39 ms) for the exponential backoffalgorithm and 160 stations (with mean packet delay of 28 ms) for the and splitting treealgorithm. Past these points, a sharp delay increase is recorded, which grows higherthan the mean inter-arrival rate (of 1s) and as a consequence mean access delaysbecome asymptotically unbounded.Mean Access Delay (ms)10009008007006005004003002001003 M bps Ups treamEthernet frames (1518-bytes )12.14 Kbps kbps per stationBackoff & Tree algorithmBackoff AlgorithmSplitting Tree Algorithm020 40 60 80 100 120 140 160 180 200 220 240 260 280 300 320 340umbe r of Active StationsFigure 5.4 – Mean access delay vs. No. of active stations, exponential backoff andsplitting tree algorithm.5-13

Chapter 5Upstream channel capacity and characterisation pIn this section and also in Section 5.4.1, we have approached the highest systemperformance when transmitting maximum-size Ethernet packets in terms of offered loadand maximum number of active stations supported. In typical networks, the packet sizesvary according to application services. Hence, in the following scenarios we addresstraffic patterns consisting mainly of Internet traffic, Voice over IP traffic, and mixedtraffic configurations.5.4.3 Effects of contention access for data transmissionAs introduced in Section 3.5, the DVB/DAVIC protocol specification supports threedifferent access modes for the transmission of data messages: contention, reservationand fixed-rate access. This section evaluates the system performance when stations areallowed to transmit small packets in the contention-based access region. The analysispresented here examines to what extent the system performance can be maximised byranging the ‘Maximum Contention Access Message Length’ of a message, measured inATM cells, that may be transmitted using contention access. Any message greater thanthis parameter is transmitted using reservation access.The analysis addresses a worst case scenario of a novel and topical mixed trafficscenario (defined below), in which most of the packets are either minimum lengthEthernet packets (in case of Internet traffic) or 40 bytes for VoIP streams. This trafficscenario was considered because small packet sizes benefit the response time ofinteractive applications. On the other hand, small packets put higher stress on thenetwork. This is because more requests per volume of data have to be issued, whichincreases the protocol overhead and the probability of collisions.The traffic load was created as follows: each active station generated 32 kbps Internettraffic as defined in Section 5.3.1. Additionally, 12.4 kbps Voice over IP traffic asintroduced in Section 5.3.2, was multiplexed into the data stream. Thus, the mean datarate per active station was set to 44.4 kbps. The simulations were performed using theexponential backoff algorithm. Figures 5.5 to 5.8 present the results when the‘Maximum Contention Access Message Length’ was ranged from 0 to 6 ATM cells.5-14

Chapter 5Upstream channel capacity and characterisation pFigure 5.5 shows that the lowest access delays can be obtained when stations onlytransmit messages comprised of 1 ATM cell in the contention access region. Low delaysunder 25 ms are produced with an offered load of 35% of the link capacity (created by24 stations), which is more suitable for the support of both Internet applications andVoIP streams. The other values (2, 3 and 5 ATM cells) also offer good results, yieldingmean access delays under 50 ms with the same traffic load. It can be appreciated that ifnone of the messages are allowed to be transmitted using contention access, higherdelays are derived, decreasing the number of supported streams (to 22 stations).100Mean Access Delay (ms)9080706050403020Contention Msg Length -- 0 ATM CellContention Msg Length -- 1 ATM CellContention Msg Length -- 2 ATM CellsContention Msg Length -- 3 ATM CellsContention Msg Length -- 5 ATM CellsContention Msg Length -- 6 ATM Cells3 Mbps Upstre am32 Kbps kbps IP &12.4 Kbps kbps VoIPBackoff A lgorithm10014 16 18 20 22 24 26umber of Active StationsFigure 5.5 – Mean access delay vs. No. of active stationsfor different contention messages.5-15

Chapter 5Upstream channel capacity and characterisation pFigure 5.6 indicates that a maximum system throughput of ≈ (1.14 Mbps) 37% of the cccan be achieved also with the value of 1 ATM cell, in comparison to ≈ (1 Mbps) 32% ofthe cc yielded when all packets are transmitted using only reservation access. Byallowing a station to transmit short packets in the contention-based region, not only afaster data transmission can be obtained (as appreciated in Figure 5.5) but also a highersystem throughput can be achieved. This is because stations do not need to waste anextra slot to send a reservation request and then wait for its acknowledgement, thusimproving the system performance. This increase in system performance can only begained in small network sizes (under ≈ 50 stations). It has also been noticed that onheavy traffic loads there is a drawback. As the ‘Maximum Contention Access MessageLength’ was increased, more data messages were transmitted using contention access.This in turn increased the collision risk with the transmissions of reservation requestssent by other stations and as a consequence system throughput is reduced. The effect isbetter appreciated in the next two figures.System Throughput (Kbps)12001140108010209609008407807206606003 Mbps Upstream32 Kbps kbps IP &12.4 Kbps kbps VoIPBackoff AlgorithmContention Msg Length -- 0 ATM CellContention Msg Length -- 1 ATM CellContention Msg Length -- 2 ATM CellsContention Msg Length -- 3 ATM CellsContention Msg Length -- 5 ATM CellsContention Msg Length -- 6 ATM Cells14 16 18 20 22 24 26 28 30 32umber of Active StationsFigure 5.6 – System throughput vs. No. of active stationsfor different contention messages.5-16

Chapter 5Upstream channel capacity and characterisation pFigure 5.7 presents the total amount of data messages transmitted in the contentionbasedregion, and Figure 5.8 shows the bandwidth consumed by collisions. Results fromFigure 5.6 revealed that with a high offered load of 37% of the link capacity (generatedby 26 stations) only 36% was successfully transmitted when using a value of 1 ATMData Transmitted in theContention-based Region (%)242118151296303 Mbps Upstream32 Kbps kbps IP &12.4 Kbps kbps VoIPBackoff A lgorithmContention Msg Length -- 0 ATM CellContention Msg Length -- 1 ATM CellContention Msg Length -- 2 ATM CellsContention Msg Length -- 3 ATM CellsContention Msg Length -- 5 ATM CellsContention Msg Length -- 6 ATM Cells14 16 18 20 22 24 26 28 30 32umbe r of Active StationsFigure 5.7 – Data transmitted in the contention-based region vs. No. of active stationsfor different contention messages.Bandwidth Consumed by Collisions0.190.170.150.130.110.090.070.050.030.013Mbps Upstream32Kbps kbps IP IP & &12.4 Kbps kbps VoIPBackoff CRAContention Msg Length -- 0 ATM CellContention Msg Length -- 1 ATM CellContention Msg Length -- 2 ATM CellsContention Msg Length -- 3 ATM CellsContention Msg Length -- 5 ATM CellsContention Msg Length -- 6 ATM Cells14 16 18 20 22 24 26 28 30 32umber of Active StationsFigure 5.8 – Bandwidth consumed by collisions vs. No. of active stationsfor different contention messages.5-17

Chapter 5Upstream channel capacity and characterisation pcell for the ‘Contention Access Message Length’ parameter (the other 1% was queued inthe station’s buffer). From this 36%, only 14.5% (Figure 5.7) of the offered load wastransmitted in contention access, and the rest (21.5%) was sent using reservation access.In order to transmit the 1-ATM-cell messages and the reservation requests of the othermessages using contention access, Figure 5.8 reveals that 10.8% of the bandwidth wasconsumed by collisions. With a value of 5-ATM cells for the same parameter, thebandwidth consumed by collisions was reported in the order of 16% of the cc, whichresults in a loss in system throughput as seen in Figure 5.6.Thus, it can be noticed that as larger messages are allowed to be transmitted usingcontention access, the risk of collision is increased slightly. In the case that none of themessages were transmitted in contention access, it is obvious that with heavy trafficloads, the bandwidth consumed by collisions is minimised.5.4.4 Effects of reservation request sizeIn this analysis, we now address the case when stations only use reservation access forthe transmission of data messages. The analysis focuses on the impact on systemperformance when the maximum request size (or ‘Maximum Reservation AccessMessage Length’ as named in the specification) is ranged from 6 to 32 ATM cells. Thisconfiguration parameter specifies the maximum length of a message that can betransmitted using a single reservation request access. Any message greater than this istransmitted by making multiple reservation requests.When a station is enabled to use only reservation access for the transmission of datamessages, as the offered load increases, short packets tend to accumulate in the station’sbuffer. This is true even at lower loads depending on the traffic type. Large packets forsome applications may generate several (or many) ATM cells at once when segmentedinto 53-bytes chunks. If a station with a large packet is allowed to send only one requestfor the transmission of the complete packet, that may affect the transmissions of theother stations. This is because the headend serves the station’s reservation requests in aFIFO order regardless of the traffic type, and the larger the request size the higher thechannel access will be, which delays the other reservation request. This effect will beaddressed at the end of this section.5-18

Chapter 5Upstream channel capacity and characterisation pIn order to demonstrate the effects of an increased reservation request size, a mixedtraffic configuration was analysed. The mean data rate per active station was set to 41.7kbps, which consisted of 32 kbps Internet and 9.7 kbps Voice over IP traffic. Thesimulations were performed using now both CRAs. In Figures 5.9 to 5.11 we comparethe results for different request limits of 6, 11, 22 and 32 ATM cells, which representthe transmission of packet sizes up to 256, 512, 1024 and 1518 bytes, respectively.Larger values were not considered, since we address the transmission of Internet andVoIP traffic that create packets under 32 ATM cells.Figure 5.9 shows that the access delays can be significantly reduced when the splittingtree algorithm is selected. For both contention resolution algorithms the lowest meanaccess delay (around 50 ms) is yielded with request sizes of 22 and 32 ATM cells,which represent the transmission of 1024 and 1518 byte packets, respectively. It can beappreciated that for these request sizes, the splitting tree algorithm supported up to 8stations more than the number supported by the exponential backoff algorithm. For ashort request size up to 6 ATM cells, this difference was of 10 stations.500Mean Access Delay (ms)450400350300250200150100500ExponentialBackoffSplittingTreeRsv Rsv Msg Msg - - 66 slotsRsv Rsv Msg Msg - - 11slotsRsv Rsv Msg Msg - - 22 22 slotsRsv Rsv Msg Msg - - 32 32 slotsRsv Rsv Msg Msg - - 66 slotsRsv Rsv Msg Msg - - 11slotsRsv Rsv Msg Msg - - 22 22 slots slotsRsv Rsv Msg Msg - - 32 32 slots3 M bps Upstream32 Kbps kbps IP IP &9.7 Kbps kbps VoIPBackoff & Tree A lgorithms20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50umbe r of Active StationsBackoffBackoffTreeTreeFigure 5.9 – Mean access delay vs. No. of active stations for different request sizes.5-19

Chapter 5Upstream channel capacity and characterisation pIn terms of system throughput, as depicted in Figure 5.10 (for large request sizes of 22and 32 ATM slots), the increase was approximately of 12% of the cc when the splittingtree algorithm was selected. Conversely, for short request sizes, the system performanceincrease can be increased up to ≈ 17%. Results for the exponential backoff algorithmrevealed that once the maximum system throughput is reached (which is between 40and 46% of the cc), as the offered load becomes higher, the throughput graduallydeclines. This is to be expected because of the increased likelihood of collisions, whichconsume more bandwidth as the cable network becomes congested.System Throughput (Kbps)20001900180017001600150014001300120011001000900800Rsv Msg - 6 slotsRsv Msg - 11slotsRsv Msg - 22 slotsRsv Msg - 32 slotsRsv Msg - 6 slotsRsv Msg - 11slotsRsv Msg - 22 slotsRsv Msg - 32 slotsBackoff Tree3 Mbps Upstream32 Kbps kbps IP & 9.7 Kbps kbps VoIPBackoff & Tree AlgorithmsSplittingTreeExponentialBackoff20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50umber of Active StationsFigure 5.10 – System throughput vs. No. of active stations for different request sizes.5-20

Chapter 5Upstream channel capacity and characterisation pThis effect can be clearly appreciated in Figure 5.11, which shows the total bandwidthconsumed by only reservation requests. For the exponential backoff algorithm, with anoffered load of 46% of the link capacity (produced by 34 stations) the bandwidthutilised by successful and unsuccessful requests ranged from 22 to 28.5% of the cc(accordingly to the request size), compared to 12.3 to 20.5% of the link capacityconsumed by the splitting tree algorithm. This difference of approximately of 8% wasaccounted for increased collisions of reservation requests produced by the exponentialbackoff algorithm, which reduces the system throughput as the offered load increases.Results presented in Figures 5.9 to 5.11 indicate that as the reservation request sizeincreases, a gain in system performance is obtained. However, increasing the requestsize to its maximum value of 32-ATM cells, (for this analysis), may have itsconsequences as shown in the following analysis.Bandwidth comsumed by reservationrequet messages (%)40363228242016128Rsv Msg - 6 slotsRsv Msg - 11slotsRsv Msg - 22 slotsRsv Msg - 32 slotsRsv Msg - 6 slotsRsv Msg - 11slotsRsv Msg - 22 slotsRsv Msg - 32 slotsBackoff Tree3 Mbps Upstream32 Kbps kbps IP & 9.7 Kbps kbps VoIPBackoff & Tree AlgorithmsExponentialBackoffSplittingTree420 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50umber of Active StationsFigure 5.11 – Reservation request bandwidth consumed vs. No. of active stations.5-21

Chapter 5Upstream channel capacity and characterisation pLet us now analyse the case when every station was transmitting either 32kbps Internettraffic or 9.7 kbps VoIP traffic. For every simulation, half of the total active stations wastransmitting Internet traffic and the other half was transmitting VoIP traffic. The packetdistribution, packet sizes and packet inter-arrival times were the same as the previousanalysis. The splitting tree algorithm was utilised for this study, since it provides abetter system performance.The main consequence of the use of the splitting tree algorithm is that with largerequest sizes (32 ATM cells), there is an increase in the access delay for Voice over IPtraffic when the offered load becomes higher than 53% of the link capacity (producedby 80 active stations). Conversely, for Internet traffic, large request sizes may result in areduction in access delay (regardless of the active number of stations) as shown inFigure 5.12.Mean Access Delay (ms)200180160140120100806040203 Mbps Upstream32 Kbps kbps IP or9.7 Kbps kbps VoIPTree AlgorithmRequest-6-cellsIP trafficRequest-32-cellsVoIPtrafficRequest-6-cells040 44 48 52 56 60 64 68 72 76 80 84 88 92 96umber of Active StationsFigure 5.12 – Mean access delay vs. No. of active stations for VoIP and IP traffic.5-22

Chapter 5Upstream channel capacity and characterisation pA recommendation can now be made for CATV networks based on the DVB/DAVICprotocol. The maximum request size should be set as large as possible if the upstreamchannel is only used for Internet traffic. A value of 32 ATM cells for this parameter wasfound to provide the best system performance. However, if the upstream channelsupports the transmission of both traffic sources (IP and VoIP), a higher interaction forVoIP streams can be obtained if the maximum request size is set as short as possible.Here a value of 6 ATM cells for this parameter was found to provide the fastestinteraction for VoIP traffic.5.4.5 Effects of buffer capacityIn most CATV networks, there are circumstances in which the externally offered load islarger than can be handled even with optimal transmissions. In these cases, if nomeasures are taken to restrict the entrance of traffic into the network, queue sizes at astation’s buffer will grow and packet delays will increase, possibly violating maximumdelay specifications. Furthermore, as queue sizes grow indefinitely, the buffer space atsome stations may be exhausted. When this happens, some packets arriving at thesestations will have to be discarded and later retransmitted, thereby wastingcommunication resources.This analysis examines to what extent traffic congestion of CATV networks can becontrolled by restricting the station buffer capacity in order to provide considerable lowdelays for the support of bursty applications (e.g. Internet traffic) and isochronousinteractive services (e.g. Voice over IP traffic, audio and video). For this analysis, amixed traffic configuration was chosen. The mean data rate per active station was set to41.7 kbps, which consisted of 32 kbps Internet traffic and 9.7 kbps VoIP traffic, asintroduced in Section 5.3.3. In order to demonstrate the effect of the buffer capacity, sixdifferent bounds were analysed (50, 100, 300, 500, 1000 and 3000 ATM cells). Thesimulations were performed using only the splitting tree algorithm.5-23

Chapter 5Upstream channel capacity and characterisation pIt can be noticed in Figure 5.13 that the access delays increase as the maximum buffercapacity builds up. The results shown in this figure indicate that the maximum numberof VoIP streams supported is 40 (which produce 54% of the link capacity), with meanaccess delays below 50 ms, regardless of the buffer size. On the other hand, bufferlimits holding up to 100 ATM cells (38 Kbits of data) yield acceptable access delays(under 1 second) for the transmission of Internet traffic, with an offered load up to ≈67% of the channel capacity, supporting up to 50 IP streams.Mean Access Delay (s)1010.10.01Buffer-50-ATM cellsBuffer-100-ATM cellsBuffer-300-ATM cellsBuffer-500-ATM cellsBuffer-1000-ATM cellsBuffer-3000-ATM cells3 Mbps Upstream32 Kbps kbps IP &9.7 Kbps kbps VoIPTree Algorithm28 30 32 34 36 38 40 42 44 46 48 50umber of Active StationsFigure 5.13 – Mean access delay vs. No. of active stations for increased buffer size.5-24

Chapter 5Upstream channel capacity and characterisation pIn terms of buffer pooling capacity, Figure 5.14 indicates that beyond an offered load of53% of the cc, the station’s buffers start building up considerably up to its maximumholding capacity. With large buffer sizes (e.g. of 1000 or 3000-ATM cells) and with anoffered load of ≈ 67% of the link capacity, a considerable amount of bandwidth (over10% of the cc) was held in the station’s buffers.We have found, on the other hand, that by using small buffer sizes (e.g. of 50 or 100-ATM cells) only a small fraction (below 1%) of the maximum channel capacity washeld in the station’s buffers. These results are more convenient for the support of VoIPstreams, because packets are transmitted faster with a small buffer size than with a largebuffer.However, having a small buffer size may results in the following drawbacks. The firstdrawback is that when a station with no available buffer space receives a packet, it willbe enabled to discard the packet, which may result in a reduction in the service quality ifthe number of discarded packets is considerably large. The second drawback if thatdiscarded packets may cause a waste of bandwidth when they are retransmitted.Therefore, it is necessary to estimate the number of packet discarded per unit of time.Mean Buffer Capacity (%)1412108642Buffer-50-ATM cellsBuffer-100-ATM cellsBuffer-300-ATM cellsBuffer-500-ATM cellsBuffer-1000-ATM cellsBuffer-3000-ATM cells3 Mbps Upstream32 Kbps kbps IP &9.7 Kbps kbps VoIPTree Algorithm045 48 50 53 55 58 61 64 67Offered Load (%)Figure 5.14 –Buffer capacity vs. offered load for increased buffer size.5-25

Chapter 5Upstream channel capacity and characterisation pFigure 5.15 shows the average number of packets that each station discarded. It can beappreciated that with an offered load beyond 53% of the link capacity, the shorter thebuffer capacity the larger the number of packets discarded, but the lower the accessdelays produced. With a very large buffer space (e.g. 3000 ATM cells) none of thepackets were discarded. However, large access delays in the order of 1 to 5 secondswere yielded, since a considerable amount of packets were held in the station’s bufferwaiting to be transmitted. Conversely, with a limited buffer capacity of 50 ATM cells,each station discards ≈ 3 packets per second.Note that when the offered load is large, low access delays and buffer overflow can becontrolled only by lowering the incoming traffic to the CATV network. Thus, there is anatural trade-off between giving sessions free access to the network and keeping delayat a level low enough so that interactive applications (e.g. VoIP, audio and video) aresupported and retransmissions or other inefficiencies do not degrade the networkperformance.Discarded packets per station (pks/s)3.53.02.52.01.51.00.50.0Buffer-50-ATM cellsBuffer-100-ATM cellsBuffer-300-ATM cellsBuffer-500-ATM cellsBuffer-1000-ATM cellsBuffer-3000-ATM cells3 Mbps Upstream32 Kbps kbps IP &9.7 kbps Kbps VoIPTree Algorithm37 39 42 45 48 50 53 55 58 61 64 67Offered Load (%)Figure 5.15 – Discarded packets vs. offered load for increased buffer size.5-26

Chapter 5Upstream channel capacity and characterisation p5.4.6 Effects of increasing the number of signalling framesAs introduced in Section 3.4, the description of upstream slots is contained in thesignalling (or MCI) field of the MPEG2 TS frame. According to the DVB/DAVICprotocol specification, these frames with signalling information should be transmitted atleast every 3 ms, when the upstream data rate is 6.176, 3.088 or 1.544 Mbps. Thefunction of these signalling frames is to co-ordinate the usage of the upstream slots,assign access modes, and indicate whether the reception of contention-based slots wassuccessful.In this section we discuss the maximum gain in system performance that can beobtained by increasing the number of MPEG2 frames with signalling information thatcan be transmitted within the 3 ms period.In a 3.088 Mbps upstream channel, the maximum number of MPEG2 frames (withsignalling information) that can be transmitted within the 3 ms period is either 1 or 2. Ina 6.176 Mbps upstream channel, this number is 1, 2 or 4. For this analysis, an Internettraffic configuration will be considered. The mean data rate per active station was set to32 kbps (as defined in Section 5.3.1). The simulations were performed using bothCRAs.5-27

Chapter 5Upstream channel capacity and characterisation pResults presented in Figure 5.16 indicate that by increasing to two the number ofMPEG2 frames (with MCI information) within the 3 ms period (represented by the label‘MCI-2’ in Figure 5.16) a reduction in access delay of approximately of 4.4 and 4.8 mscan be obtained for the exponential backoff algorithm and the splitting tree algorithm,respectively. This reduction can only be obtained with traffic loads under 47% (45active users) and 53% (51 active users) of the cc for the exponential backoff algorithmand splitting tree algorithm, respectively.However, for high traffic loads, lower access delays are produced with one signallingframe in every 3 ms period (represented by ‘MCI-1’). This change in access delay isbecause as the network load becomes higher than the point of saturation, there is anincreased risk of collision of reservation requests with ‘MCI-2’ (according to simulationresults), which degrades slightly the overall system performance as shown in Figure5.16.Mean Access Delay (ms)1201101009080706050403020100Backoff-MCI-1Backoff-MCI-2Tree-MCI-1Tree-MCI-23 Mbps Upstream32 Kbps kbps IPBackoff & Tree Algorithm30 33 36 39 42 45 48 51 54 57 60 63 66umber of Active StationsFigure 5.16 – Mean access delay vs. No. of active stations for different MCI values.5-28

Chapter 5Upstream channel capacity and characterisation pOn the other hand, a loss of 6% in terms of system throughput was seen with theexponential backoff algorithm for ‘MCI-2’, as depicted in Figure 5.17. The splitting treealgorithm did not provide a loss for ‘MCI-2’, because this algorithm resolves collisionsmore efficiently than the exponential backoff algorithm, regardless of the number ofMPEG2 frames (with MCI information) within the 3 ms period. This inefficiency of theexponential backoff algorithm results in a reduction in system throughput.System Throughput (%)66636057545148454239363330Backoff-MCI-1Backoff-MCI-2Tree-MCI-1Tree-MCI-23 Mbps Upstream32 Kbps kbps IPBackoff & Tree Algorithm30 33 36 39 42 45 48 51 54 57 60 63 66umber of Active StationsFigure 5.17 – System throughput vs. No. of active stations for different MCI values.5.4.7 Effects of packet size for single and multiple node scenariosAnother critical parameter in the performance of any computer network is the packetsize. From results in Section 4.5, packet size was proved by both analytical model andsimulation (Figure 4.18) to have a significant impact on system performance, especiallywhen small and large packets are considered for data transmissions.5-29

Chapter 5Upstream channel capacity and characterisation p5.4.7.1 Single node scenarioThis analysis addresses the issue of maximum throughput per station that thecommunication protocol can achieve. For this particular case, a single node scenarionetwork is considered and simulations were run for different Ethernet packet sizes (from64 to 1518 bytes). Tests were performed with the exponential backoff algorithm.Results for the splitting tree algorithm reported the same system performance, becausewith only one station collisions are avoided. In order to obtain the maximum achievablethroughput, the station was set to produce the maximum number of packets that it couldreceive from an Ethernet network, which was a 10 Mbps constant stream with variablenumber of packets per second depending on the packet size.The system throughput and upstream channel utilisation versus the packet sizegenerated is shown in Figure 5.18. The results revealed that the communicationprotocol is highly inefficient. The maximum throughput does not exceed 32% (=994kbps) of the link capacity using maximum Ethernet packet size, even though the offeredload was a constant 10 Mbps stream.In this analysis it was found that the scheduler-look ahead is the major delay element inthe transmission cycle, which dramatically reduces the throughput.Throughput & Channel Utilisation (%)504540353025201510503 Mbps Upstream10 Mbps Ethernet TrafficBackoff AlgorithmUtilisationThroughput64 128 256 512 768 1024 1280 1518Packet Size (bytes)Figure 5.18 – System throughput/utilisation vs. packet size.5-30

Chapter 5Upstream channel capacity and characterisation pFor example, in the worst-case scenario, when packet sizes of 64 bytes are transmitted,the maximum throughput that a single node can achieve resulted as low as 56 kbps,which corresponds to 1.84% of a 3.088 Mbps upstream channel.A comparison between the throughput in kbps and the throughput in packets per secondis presented in Figure 5.19. The increase in packet size compensates for the decrease ofthroughput in terms of packets per second. We can appreciate that after a 768-bytepacket size, there is a considerable fall in the number of packets transmitted per second.This is reasonable, since the signalling frame only describes up to 18 slots per 3 msperiod, and when all slots are used for data transmission, a packet size up to18 Slots ⋅ 48bytes_ payload = 864bytescan be transmitted in one signalling frame. Larger packet sizes require two signallingframes. Regardless of the packet size, two more signalling frames are required forcontention access, which results in 1 /(3MCI frames⋅ 3ms)= 111 frames transmitted persecond for packet sized under 864 bytes (1.8% deviation for simulation results) and1/4·3ms = 83 frames transmitted per second for packet sized higher than 864 bytes(1.2% deviation for simulation results).1000110900Throughput packets/s10080090Throughput (Kbps)700600500400300200100Throughput Kbps3 Mbps Upstream10 Mbps Ethernet TrafficBackoff Algorithm8070605040302010Throughput (packets/s)064 128 256 512 768 1024 1280 1518Frame Size (bytes)0Figure 5.19 – System throughput in kbps and packets/s vs. packet size.5-31

Chapter 5Upstream channel capacity and characterisation pAs a final remark for this analysis, it was seen that the reservation request mechanism ofthe DVB/DAVIC protocol was not optimised, since a poor performance is obtainedwhen short messages are being transmitted. Therefore, it is evident that a performanceoptimisation or the incorporation of new mechanisms should be considered.5.4.7.2 Multiple node scenarioIn this section, we now consider a multiple-node network scenario to study themaximum performance gain that can be obtained by increasing the packet size. Here,isochronous streams of 32 kbps (produced by each station) were analysed and the interarrivalrates were set according to the packet size. The simulations were carried outusing only the splitting tree algorithm because of its increased performance over theexponential backoff algorithm. The packet sizes considered were 64, 128, 256, 512,1024 and 1518 bytes that cover the whole range from the minimum to the maximumEthernet packet size.Figures 20 to 22 present a general performance overview when the packet size isincreased. The delay and throughput results revealed that for small packet sizes, thecommunication protocol is also highly inefficient.System Throughput (%)7065Pk_Size: 64 bytesPk_Size: 128 bytes60 Pk_Size: 256 bytes55Pk_Size: 512 bytesPk_Size: 1024 bytes50 Pk_Size: 1518 bytes45403530253 Mbps Upstream2032 kbps Kbps isochronous Isochrous streamsTree Algorithm1518 21 24 27 30 33 36 39 42 45 48 51 54 57 60 63 66 69umber of Active StationsFigure 5.20 – System throughput vs. No. of active stations for increased packet size.5-32

Chapter 5Upstream channel capacity and characterisation pThe maximum throughput (see Figure 5.20) yielded for minimum Ethernet packet sizewas as low as ≈ 34% of the link capacity, which can only support up to 25 stationsbefore delays became extremely large for the delivery of isochronous streams (as shownin Figure 5.21). Increasing the packet size can significantly improve the performancewith maximum sustainable throughput up to ≈ 64% of the cc. This increase in systemperformance was recorded for packet sizes of 512 bytes. Larger packet sizes have onlymarginal increase in system throughput and a small increase in packet access delay.From Figure 5.21, for offered loads less than 47% of the cc (produced by up to 45stations), access delays for short packet sizes (under 1024 bytes with the exemption ofthe 64-byte packet size) were relatively shorter than for larger packet sizes (e.g. 1024and 1518 bytes). This is mainly due to the larger transmission delay incurred by longpackets. On average, when a station has gained access to the reservation access region,it takes 1 signalling frame of 3 ms to transmit a packet up to 512 bytes and 2 signallingframes to transmit a packet of 1024 or 1518 bytes. Hence, a minimum difference of 3ms between short and large packet sizes can be noticed in Figure 5.21.Mean Access Delay (ms)504540353025201510503 Mbps Upstream32 32 kbps Kbps isochronous Isochronous streamsTree AlgorithmPk_Size: 64 bytesPk_Size: 256 bytesPk_Size: 1024 bytesPk_Size: 128 bytesPk_Size: 512 bytesPk_Size: 1518 bytes18 21 24 27 30 33 36 39 42 45 48 51 54 57 60 63 66 69umber of Active StationsFigure 5.21 – Mean access delay vs. No. of active stations for increased packet size.5-33

Chapter 5Upstream channel capacity and characterisation pHowever, with higher offered loads (above ≈ 47% of the cc), short packets cause morerequests to be sent in the contention access region, resulting in more collisions (lessthroughput). This effect is better appreciated in Figure 5.22. This figure shows thepercent of the bandwidth wasted by reservation requests, which resulted in collision. Itis evident that with a 64-byte packet size, as the offered load increases, a considerableamount of bandwidth is wasted. For instance, with offered loads over 30% of the cc,approximately up to 8% was consumed only by collisions, which reduced considerablythe bandwidth available for data packets. As a result of this reduction in the reservationaccess region, the stations are unable to transmit all data packets due to the high risk ofcollision among reservation request, which cause the station’s buffers to build up almostunbounded. Thus, very large packet access delays are evident with an offered load aslow as 28% of the cc (produced by 27 stations) as indicated in Figure 5.21.With a 128-byte packet size, there was also a sizeable amount (up to ≈ 4%) wasted bycollisions. However, with large packet sizes, (e.g. 1024 and 1518 bytes) the bandwidthwasted by collisions only represented a tiny portion of the link capacity (under 0.05% ofBandwidth Comsumed by CollidedContention Slots (%)987654321Pk_Size : 64 bytesPk_Size : 128 bytesPk_Size : 256 bytesPk_Size : 512 bytesPk_Size : 1024 byte sPk_Size : 1518 byte s3 Mbps Upstream32 Kbps kbps isochronous Isochrous streamsTree Algorithm018 21 24 27 30 33 36 39 42 45 48 51 54 57 60 63 66 69umbe r of Active StationsFigure 5.22 – Bandwidth consumed by collisions vs. No. of active stationsfor increased packet size.5-34

Chapter 5Upstream channel capacity and characterisation pthe cc).As a final analysis, Table 5.5 presents a summary of the maximum system performanceand bandwidth characterisation for the six different packet sizes considered. In this tablewe can appreciate that regardless of the packet size, the maximum system utilisation isaround 89-90% of the link capacity, with a maximum deviation betweenthroughput/utilisation of ≈ 55% and ≈ 24% for minimum and maximum Ethernet packetsizes, respectively. There are several factors that may account for these deviations.A major component is due to the relatively large amount of overhead involved at MAC(ATM encapsulation) and PHY (forward error correction and synchronisation) layers.This can be as large as 34% and 23% of the cc when delivering small and large packetsof 64 and 1518 bytes, respectively.Another factor is the amount of bandwidth used for the transmission of successfulreservation requests. For 64-byte packets, this amount of bandwidth (of ≈ 12%)represents a significant portion of cc. However, for 1518-byte packets, this bandwidth(of ≈ 1%) represents a tiny fraction of the link capacity. This difference is because shortmessages require proportionally more reservation requests in order to transmit the sameamount of data than with larger messages. For instance to transmit 32 kbps, a stationwill send 75 requests per second with a 64-byte packet size, in comparison to 2.64requests per second with a 1518-byte packet size.Table 5.5 - Maximum system performance and bandwidth characterisation.BandwidthPacket size (bytes)Characterisation (%) 64 128 256 512 1024 1518Maximum throughput 34.5 51.3 56.2 62.9 63.4 64.6Maximum utilisation 89.1 90.1 90.5 89.8 88.9 88.5Deviation 54.6 38.8 34.3 26.9 25.5 23.9Supported streams 25.0 48.0 54.0 60.0 60.0 63.0Request collided 8.3 4.1 1.3 0.4 0.1 0.1Request successful 11.8 8.5 4.6 2.6 1.3 0.9MAC&PHY overhead 34.5 26.2 28.4 23.9 24.1 23.0Bandwidth unused 10.9 9.9 9.5 10.2 11.1 11.55-35

Chapter 5Upstream channel capacity and characterisation pThe next significant factor, as stated earlier, is the bandwidth wasted by unsuccessful(or collided) reservation request, which can result in a significant amount when shortmessages are delivered.Finally, the other obvious performance issue is the inability of the system to reach themaximum channel capacity (≈ 100%). The reason for this is the minimum number ofcontention slots allocated per signalling frame, which was set to 2 in this scenario (andaccounted for ≈11% cc).For 1518-byte packets, some additional CSs were added to the signalling frame becauseof the inability of the headend to allocate all reservation slots for the transmission ofdata packets. Hence, the total bandwidth allocated to the contention access region(registered by simulations) accounted for ≈12% of the link capacity, where only ≈ 1%was used for successful and unsuccessful requests and the other bandwidth, of 11%,remained unused.In general, the unused bandwidth can be reduced by decreasing the number of CSsallocated per signalling frame. However, this may be inefficient for short packet sizes,and particularly when the exponential backoff algorithm is used, as examined in theSections 5.4.2 and 5.4.4.We have also carried out a detailed performance analysis of isochronous streams (at8kbps, 16 kbps, 32 kbps, 64 kbps and 128 kbps) for a 6 Mbps upstream channel and sixdifferent packet sizes (64, 128, 256, 512, 1024 and 1518 bytes). This analysis isreported in [85] and presented in Figure E.2 of Appendix E.5.5 ConclusionsIn this chapter, four novel and topical traffic types were used to analyse the fundamentalperformance characteristics of DVB/DAVIC compliant networks with the use of asimulation model. This data facilitates further optimisations and enhancements in theprotocol development of DVB/DAVIC. We studied key issues on system performanceusing both contention resolution algorithms, such as: packet access delay and maximumthroughput that can be achieved by a single station; packet size scalability, optimisation5-36

Chapter 5Upstream channel capacity and characterisation pof the maximum length of a message (measured in ATM cells) that can be transmittedusing either contention access or a single reservation request; buffer capacity; scalabilityof the upstream channel in terms of increased offered load and node population.The major factors affecting the system performance were seen to be the length of thepacket being transmitted for delivery. Specifically it was demonstrated that regardless ofthe offered load and length of the packet size, a single station cannot achieve throughputhigher than 32% of the maximum upstream channel capacity, even when the splittingtree algorithm is used. However, this figure can be as low as 1.8% when deliveringminimum Ethernet packets (64 bytes) due to an excessive number of reservationrequests, collisions, retransmissions and DVB MAC and PHY protocol overheads.Results for channel utilisation were higher than for throughput results with a range of54% to 23% for minimum and maximum Ethernet packets, respectively. Protocoloverhead was shown to be a major source of performance inefficiency. This can be ashigh as 35% of the upstream channel capacity for 64-byte packets.Effects of contention access for data transmission revealed that by transmittingrelatively short messages (in the order of 1-ATM cell) using only contention access, thesystem throughput can be increased and a quick interaction can be provided. Results foreffects of reservation request size suggested that the maximum request size should be setas large as possible if the upstream channel is only used for Internet traffic. However, ifthe upstream channel supports the transmission of both traffic types (IP and VoIP), ahigher interaction for VoIP packet can be obtained if the maximum request size is set asshort as possible.Results for the buffer capacity indicated that traffic congestion could be controlled bylowering the incoming traffic to the CATV network. Thus, there is a natural trade-offbetween giving sessions free access to the network and keeping delay at a level lowenough so that interactive applications (e.g. VoIP, audio and Video) are supported andretransmission or other inefficiencies do not degrade the network performance.Scalability of the upstream channel revealed that there is a gradual increase in theperformance of the system with respect to increasing the cable population and offeredload. Results also revealed that there are distinct saturation points after which5-37

Chapter 5Upstream channel capacity and characterisation pthroughput and utilisation do not increase, meanwhile packet delays becomeunbounded. For all of these analyses, the maximum channel utilisation remained atapproximately 90% of the link capacity.Careful design should not allow a network population and offered load to exceedsaturation points, since this would result in uncontrolled access delay.Finally, given all these special characteristics of the protocol and the fact thatisochronous applications (e.g. VoIP) are very sensitive to delay variations, it isnecessary to further optimise the protocol.5-38

Chapter 6OPTIMISATION OF CRAALGORITHMS USINGADAPTIVE CSA6.1 IntroductionContention resolution algorithms (CRA) define the set of rules used to resolvecollisions. They play a vital role in the performance of a multi-access reservationprotocol. This is because the faster they resolve collisions, the lower the access delaywill be and the higher the system throughput will become. Initially, CRAs gained muchinterest in the early 1970s for usage in the transmission of radio packets, and especiallyduring the development of the ALOHANET project [1]. Two major candidates weredefined, ALOHA-based algorithms like ‘exponential backoff and p-persistence’ andsplitting tree algorithm, as reported in Section 1.2.3.2. Since then, much research hasbeen devoted to devising efficient contention resolution mechanisms for multi-accessmedia for Local Area Networks (LANs), Metropolitan Area Networks (MAN), satellitenetworks, radio networks and CATV networks [41].In this chapter, we study in detail performance, optimisation and implementation issuesfor the contention resolution algorithm adopted by the DVB/DAVIC protocol. We payparticular attention to the dynamics and operation of the exponential backoff algorithmand the splitting tree algorithm. Furthermore, special emphasis is paid to the design ofadaptive mechanisms, called Contention Slot Allocators (CSA), which dynamicallyadjust the bandwidth used for contention access, significantly increasing the systemperformance when different bounds are considered.6-1

Chapter 6Optimisation of CRA using adaptive CSAs p6.2 Contention Slot AllocatorAs introduced in [61], [96] and [98], the authors have pointed out that the performanceof a multi-access reservation protocol depends more on the overall framing structureand the capacity assigned to the contention and reservation access modes than thedetails of the CRA adopted. In this section we focus on the performance impact whenthe reservation capacity is dynamically adjusted by the use of a slot allocationmechanism.6.2.1 How many contention-slots per signalling frame?After the INA has scheduled a number of reservation slots (RSs) to carry data packets,any number of contention slots (CSs) may then be allocated. When the load of thenetworks is low, very few CSs are required. On the other hand, since the load is low,there will be unused slots that could be used as CSs. As the offered load increases,depending on the length of the packets, more slots will need to be allocated as CSs.The solution to determine how many CSs is rather simple: allocate all slots that are notbeing used for data as CSs. At low traffic loads, many more CSs will be allocated thanthose required. The surplus of CSs significantly decreases the risk of collision ofreservation requests to a very low level, which in turn reduces the access delay for datapackets.This algorithm is a self-regulating mechanism, since if the number of CSs are too low,requests will not reach the INA and as a result more CSs will be automatically allocated.On the contrary, if the number of CSs is too high, more successful requests will reachthe INA and the number of empty slots that can be used as CSs will decrease to aminimum threshold value, which guarantees that at least a few slots will be reserved forcontention access. Thus, the performance of the network is highly dependant on theminimum number of CSs allocated in each signalling frame. In [61], [96] and [98] theauthors did not consider the minimum number of contention slots that should beallocated in each signalling frame, since this would have led them to a low performanceestimation. In the next section, we present a performance analysis for different valuesof the ‘Minimum number of CSs per signalling frame’ parameter using both CRAs.6-2

Chapter 6Optimisation of CRA using adaptive CSAs p6.2.2 Simple-CSA: Performance optimisation of the use of contentionslotsIn order to demonstrate the effects of the use of a simple contention slot allocator, anInternet traffic situation was analysed. The mean data rate per active station was set to64 kbps. The simulations were performed using both contention resolution algorithms.The Min. o. of CSs per signalling frame was ranged from 1 to 7 CSs for theexponential backoff algorithm. For the splitting tree algorithm this parameter wasranged from 0 to 6 CSs.6.2.2.1 Exponential backoff algorithm performance using a Simple-CSAWe have used Equation 4.43, (introduced in Section 4.4.3) to estimate the maximumupstream throughput for the exponential backoff algorithm.Smax_ exp_ backoffPksize=Pk + eCSsslotsThe term ‘eCSs’ in this equation, indicates that on average we need ‘e(=2.718)’contention slots to transmit a reservation request successfully, in order to get theoptimum system throughput, as suggested in [10]. Idles occur with a probability of 1/e ≈0.368, success occurs with a probability of 1/e and collisions occur with a probability of1-2/e ≈ 0.264. Thus in summary, 1 CSs for idle, 1 CSs for success and 0.718 CSs forcollisions are needed.On average the packet size for Internet traffic is 368.1 bytes [53], and the number ofdata slots requested per packet is 8.3. So using Equation 4.43 we have,Smax368.1== 52.2% ≈ 52%8.3*64 + 2.718*64Therefore, the maximum theoretical system throughput that can be achieved is ≈ 52%when large numbers of stations are transmitting.6-3

Chapter 6Optimisation of CRA using adaptive CSAs pFrom simulation results presented in Figures 6.1 and 6.2, it is apparent that with at least3 or 4 CSs per signalling frame (represented by CSs-3 and CSs-4 in these figures), notonly the lowest access delays are yielded, but also the highest system throughput isachieved.Mean Access Delay (ms)50045040035030025020015010050CSs -1CSs -2CSs -3CSs -4CSs -5CSs -6CSs -73 Mbps Upstream64 Kbps kbps IPBackoff CRA018 19 20 21 22 23 24 25umbe r of Active StationsFigure 6.1 – Mean access delay vs. No. of active stations.Exponential backoff algorithm with a Simple-CSA.System Throughput (%)52CSs -150CSs -248CSs -3CSs -446CSs -5CSs -644CSs -742403 M bps Upstream3864 Kbps kbps IPBackoff CRA3618 19 20 21 22 23 24 25 26 27 28 29 30umbe r of Active StationsFigure 6.2 – System throughput vs. No. of active stations.Exponential backoff algorithm with a Simple-CSA.6-4

Chapter 6Optimisation of CRA using adaptive CSAs pLarger values, (above 4 CSs per signalling frame) cause a waste of bandwidth, since theCSA tends to allocate more contention slots than those needed. This in turn results in areduction in system throughput. Furthermore, this causes increased access delay asstations transmit reservation requests more frequently. Such requests at the headendhave to wait longer for grant since there is a reduction in the bandwidth assigned forreservation access.From Figure 6.3, it is evident that on high traffic offered loads, beyond 50% of the cc,(generated by at least 24 stations), the average number of contention slots needed perrequest ranged from 3 to 3.5 slots, when the minimum number of CSs per signallingframe was below 5 slots.In Figure 6.2 we can appreciate that the maximum system throughput achieved wasapproximately 51% (for CSs-4) with an offered load about 52% of the channel capacityyielded by 25 stations. This difference of approximately 1% between theoretical andsimulation results was attributed to the fact that the CSA assigned on average 3.08contention slots per successful request (see Figure 6.3), instead of the optimum value of2.718. Furthermore, the average Pk size (registered by simulations) was 374 bytes with8.42 slots requested per packet, which made the difference of 1%.Average umber ofContention Slots per Request7.57.06.56.05.55.04.54.03.53.02.5CSs-1CSs-2CSs-3CSs-4CSs-5CSs-6CSs-73 Mbps Upstream64 Kbps kbps IPBackoff CRA18 19 20 21 22 23 24 25 26 27 28 29 30umber of Active StationsFigure 6.3 – Average No. of CSs per reservation request vs. No. of active stations.Exponential backoff algorithm with a Simple-CSA.6-5

Chapter 6Optimisation of CRA using adaptive CSAs p6.2.2.2 Splitting tree algorithm performance using a Simple-CSAFor this algorithm, the maximum theoretical bound for the system throughput can nowbe obtained by using Equation 4.44, in which reservation requests are transmitted usingminislots (MSs) of 21 bytes. The maximum theoretical system throughput that can beachieved with the splitting tree algorithm is ≈ 63% of the channel capacity, as shownbelow.SPk368.1== 62.6% ≈ 63+ eMSs 8.3*64 + 2.718* 21max=%PkslotsFrom simulation results shown in Figures 6.4 to 6.6, it is evident that the highest systemperformance is yielded with at least 1 CSs per signalling frame. This performance isachieved because a contention slot is reserved in the next signalling frame when acollision happens. This slot (which will be divided into three minislots) is then usedamong stations that caused the collision. By reserving an additional contention slot aftera collision, apart from the Min. o. of CSs per signalling frame, the Contention-Resolution-Grant Cycle (described in Section 3.4.1) is considerably decreased, thusresulting in a reduction in access delay.Mean Access Delay (ms)50045040035030025020015010050CSs-0CSs-1CSs-2CSs-3CSs-4CSs-5CSs-63 Mbps Upstream64 Kbps kbps IPTree CRA018 19 20 21 22 23 24 25 26 27 28 29 30 31umber of Active StationsFigure 6.4 – Mean access delay vs. No. of active stations.Splitting tree algorithm with a Simple-CSA.6-6

Chapter 6Optimisation of CRA using adaptive CSAs pHaving a short value for the Min. o. of CSs per signalling frame, (e.g. 1 or even 2CSs), most of the bandwidth is allocated to the reservation access region, producing anincrease in the overall system performance. Larger values for this parameter (above 2CSs) are not required, because this algorithm efficiently resolves collisions. Defining alarge value for this parameter may result in many contention slots being unused, wastingbandwidth resources and therefore achieving a low system performance.Conversely, having at least 0 CSs per signalling frame may cause (on heavy loads)those stations with new incoming packets to wait until all the previous reservationrequests have been granted. Then, compete for contention access among all waitingstations. This increases considerably both the risk of collisions and the CRGC, resultingalso in a reduced performance.From Figure 6.5 the maximum system throughput obtained by using this algorithm wasapproximately 62%. The remaining bandwidth of 0.6%, (to get the 62.6% as estimated)was attributed, in this case, because the mean Pk size (registered by simulations) was of374 bytes with 8.42 slots requested per packet.System Throughput (%)6562595653504744413835CSs-0CSs-1CSs-2CSs-3CSs-4CSs-5CSs-63 Mbps Upstream64 Kbps kbps IPTree CRA18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33umber of Active StationsFigure 6.5 – System throughput vs. No. of active stations.Splitting tree algorithm with a Simple-CSA.6-7

Chapter 6Optimisation of CRA using adaptive CSAs pIn addition, the CSA assigned on average 1.07 CSs (or 3.21 MSs) per successful requestas shown in Figure 6.6, instead of the optimum value of ‘2.718 MSs ≈ 0.89 CSs’.As a general observation, results presented for both CRAs showed that the optimalsystem performance depends on the Min. o. of CSs per signalling frame allocated. Forthe exponential backoff algorithm we found that the optimum value for this parameterwas 4 CSs for a maximum system throughput and 3 CSs for lower access delays. Forthe splitting tree algorithm this value appeared to be of 1 CSs.Average umber of ContentionSlots per Request7.57.03 Mbps Upstream6.564 Kbps kbps IPTree CRA6.05.55.04.54.0 CSs-03.5 CSs-13.0 CSs-22.5 CSs-32.0 CSs-41.5 CSs-51.0 CSs-60.518 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33umber of Active StationsFigure 6.6 – Average No. of CSs per reservation request vs. active stations.Splitting tree algorithm with a Simple-CSA.6-8

Chapter 6Optimisation of CRA using adaptive CSAs pHowever, for different traffic types (e.g. Internet at 32 kbps per station and mixed at41.7 kbps traffic as summarised in Table 6.1), for the exponential backoff algorithm, theoptimal system performance, was found with at least 4 and 5 CSs per signalling framefor Internet and mixed traffic, respectively. For the splitting tree algorithm this value(1CSs) was unchanged. In the following section we introduce two enhanced CSAs,which will optimise further the maximum system performance.Table 6.1 - Maximum system performance for Internet and mixed traffic scenarios.CRAExponentialBackoffAlgorithmTrafficTypeInternetInternetInternetInternetMixedMixedMixedMixedMin.CSs perSign. FrameMaximumThroughput (%)Mean AccessDelay (ms)CSs perRequestOfferedLoad (%)ActiveStations2 47.9 236 3.6 48.4 483 47.9 602 3.62 49.4 484 49.3 57 3.31 49.4 485 48.4 178 3.51 48.8 482 43.5 1435 3.26 45.7 343 43.8 926 3.18 45.7 344 44.5 764 3.05 46.1 345 45.23 179 2.88 45.5 34SplittingTreeAlgorithmInternet 1 61.01 155 1.09 61.1 60Internet 2 58.06 189 1.54 58.5 57Internet 3 55.07 153 2.08 55.1 54Internet 4 51.6 257 2.8 52.2 51Mixed1 57.27 282 0.93 57.9 43Mixed2 55.3 98 1.27 55.5 41Mixed3 52.4 331 1.63 53 39Mixed 4 48.99 191 2.15 49.4 37Shaded rows represent the configuration for optimum system performance.6.2.3 Enhanced - CSAsIn this section two enhanced CSAs are introduced. We have called these enhancedmechanisms ‘Forced-CSA’ and ‘Variable-CSA’. Such mechanisms adjust dynamicallythe number of CSs per signalling frame according to the current traffic load, meanpacket size, mean requested slots and possible collisions. These mechanisms willimprove the maximum system performance for the exponential backoff algorithm bysending more CSs when they are needed (and not when they are available) and byreducing the average number of contention slots needed per request, from 3.08 CSs(Figure 6.3) to a value very close to the optimum ‘e(=2.718)’6-9

Chapter 6Optimisation of CRA using adaptive CSAs pFor the splitting tree algorithm a more efficient CSA is not necessary, since the systemperformance has already been maximised by fixing the Min. o. of CSs per signallingframe to 1 CSs, regardless of the traffic type and when a medium size network has beenconsidered (under 70 stations).6.2.3.1 Forced-CSA used with the exponential backoff algorithmThis mechanism is based on the dynamics of the splitting tree algorithm. When acollision occurs, the splittting tree algorithm automatically allocates a CS in the nextsignalling frame, which is then split into 3MSs and used only between the stationsinvolved in the collision. The Forced-CSA, on the other hand, allocates a flexiblenumber of CSs in the next signalling frame. We refer to these additional slots as forcedcontention slots (FSs). With this new functionally, stations competing for contentionaccess have more chance to transmit successfully, since more contention slots areallocated when they are needed, reducing considerably packet access delays.The idea of allocating more contention slots, in additional to the unreserved slots thatare also allocated to the contention-based access region, was first reported in [96] and[98]. Here the authors introduced a new contention slots allocator, referred to as‘Forced Mini-Slots CSA’ for the IEEE 802.14 protocol. The main difference betweenour Forced-CSA and the Forced Mini-Slots CSA, is that the latter allocates more CSsaccording to the maximum throughput of the Slotted Aloha System, defined inEquation 6.1. a −1ηmax= a⋅ p ⋅ (1 − p)(6.1)where p is the retransmission probability and a is an estimation of the number ofstations competing for a contention slot.On high traffic loads, the Forced Mini-Slots CSA allocates e(=1/η max ) CSs for each datamessage to be transmitted. Conversely on low traffic loads, it allocates less thane(=1/η max ) CSs. However, our mechanism (Forced-CSA) allocates more CSs when acollision occurs and on high traffic loads, the average number of CSs required perrequested data message approaches very close to e(=1/η max ).6-10

Chapter 6Optimisation of CRA using adaptive CSAs pFor the Forced Mini-Slots CSA, the authors found that by setting the number of forcedmini-slots to 2 instead of e, good results were obtained. In our analysis, we also foundthat by allocating 2 FSs, after a collision, an improvement in system performance wasobtained. Results for the Forced-CSA are presented in Section 6.2.3.3.6.2.3.2 Variable-CSA used with the exponential backoff algorithmThis mechanism uses a variable slot regime algorithm in which the ratio of CSs toreservation data slots (RSs) is varied from signalling frame to signalling frame, based onthe current traffic load, mean packet size and the mean of requested slots. Variable slotallocators have been used since 1998 in MAC protocols for HFC and wireless accessnetworks. They were firstly introduced in [112] and [113]. Later on, they were reportedin [41] for the IEEE 802.14 protocol. The mechanism presented here is similar to [41]with slight modifications for the DVB/DAVIC protocol. The variable number of CSs(VCSs) to be added in the signalling frame is dynamically adjusted as the headendconverts the number of RSs into CSs (N RSs) according to Equation 6.2. RSs=⎡2⋅⎢⎣ 2MAX+ Re_ DATA⎤q _ Size⎥⎦(6.2)where MAX_DATA is the maximum number of data slots that can fit in a signallingframe (= 18 – Min. o. of CSs per signalling frame, for a 3.088Mbps upstreamchannel) and Req_Size is the average number of RSs that can be requested at once.VCSs can be determined from Equation 6.3.⎧0if MAX_REQ ≥ α ⋅ (MAX_DATA - NRSs)VCSs = ⎨(6.3)⎩NRSselsewhere MAX_REQ is the total number of data slots requested but not yet allocated bythe headend, α is a design parameter set to 2.5 as suggested in [41]. The total number ofCSs to be included in the next signalling frame is then represented by followingparameters: the Min. o. of CSs per signalling frame (as proposed in the Simple-CSA),the variable number of CSs derived from Equation 6.3 and the unused RSs converted toCSs as recommended in [96] and [98] .6-11

Chapter 6Optimisation of CRA using adaptive CSAs p6.2.3.3 Performance comparison of Enhanced-CSAsIn order to provide a complete performance comparison between the Forced andVariable CSAs, three different traffic configurations will be analysed: 64 kbps IPstreams, 32 kbps IP streams and 41.7 kbps mixed traffic.A summary of the maximum system performance produced by each CSA for Internettraffic is presented in Table 6.2. A detailed comparison for mixed traffic is shown inFigures 6.7 to 6.10.For Internet traffic at 32 kbps per station, the best system performance was achieved bythe Forced-CSA. Three different configurations were analysed using this scheme (asindicated in Table 6.2). The highest system throughput (about 52% of the cc) and thelowest mean access delay (of 584 ms) was achieved by both allocating 2 FSs after acollision and defining at least 2 CSs per signalling frame, at approximately 53% of thechannel capacity (produced by 51 stations). In addition, the average number of CSsneeded per request resulted in 2.819, which is quite close to the optimum e(=2.718) witha maximum deviation under 1% of the cc between simulation and theoretical results.Table 6.2 – Maximum system performance for Internet traffic and different CSAs.InternetTraffic32 kbpsCSASimpleSimpleMin. CSsPer Sign.FrameMaximumThroughput(%)Mean AccessDelay(ms)CSs perRequestOfferedLoad(%)ActiveStations3 46.8 3099.98 3.96344 52.3501 515 48.0 2004.82 3.57891 51.3815 51Forced-FSs 1 3 51.1 1284.42 2.87358 53.1611 51Forced-FSs 2Forced-FSs 2VariableVariable2 51.5 584.026 2.81989 52.8379 513 51.0 1090.09 2.91893 53.0158 512 47.9 2628.13 3.68146 53.153 513 47.8 2195.96 3.6845 52.9651 5164 kbpsSimple 4 50.6 1226.38 3.07588 52.5 25Forced-FSs 1 3 50.7 913.49 3.01059 51.7 25Forced-FSs 2 2 50.6 494.56 3.01297 51.3 25Forced-FSs 2 3 49.8 416.05 3.15268 50.7 25Variable 2 50.8 777.72 2.95227 51.9 25Variable 3 51.3 624.79 2.87162 52.1 25The shaded rows present the CSA with optimum system performance.6-12

Chapter 6Optimisation of CRA using adaptive CSAs pFor Internet traffic at 64 kbps per station, the highest system throughput was achievedwith the Variable-CSA and the lowest access delay with the Forced-CSA. For theVariable-CSA two configurations were used. The maximum system throughput ofapproximately 51% at 52% of the channel capacity (generated by 25 stations) was foundby allocating at least 3 CSs per signalling frame. For this traffic type, a slight increase inthe average number of CSs needed per request (with 2.871 CSs) was registered, whichresulted in a small reduction in system performance of 0.2% in comparison with Internettraffic at 32 kbps.For a better appreciation of the performance impact when using enhanced CSAs, amixed traffic situation will be carefully analysed. For this traffic type (comprised of 32kbps Internet and 9.7 kbps VoIP traffic), the maximum theoretical bound for the systemthroughput can now be estimated by taking the mean packet size transmitted (measuredin bytes and in slots), as indicated in Equation 6.4.PkPbIP⋅ PkIP+ PbVoIP⋅ PkVoIPSmax ==(6.4)Pk + eCSs Pb ⋅ Pk + Pb ⋅ Pk eCSsslots IP slots_IP VoIP slots_VoIP+where Pk IP (=368.1) and Pk slots_IP (8.3) are the average Internet packet sizes measured inbytes and data slots respectively. Similarly, Pk VoIP (=146) and Pk slots_VoIP (=4) are theVoIP packet sizes measured in bytes and data slots respectively. Finally, Pb IP and Pb VoIPare the probability that an IP or VoIP packet will be generated, respectively. Equations6.5 and 6.6 give these probabilities.PbPbDrIPIP= (6.5)PkIPDrIP+ ⋅ DrVoIPPkVoIPDrVoIPVoIP= (6.6)PkVoIPDrVoIP+ ⋅ DrIPPkIPwhere Dr IP (= 32 kbps) and Dr VoIP (= 9.7 kbps) are the mean data rates of Internet andVoIP traffic generated per station, respectively. Therefore, the maximum theoreticalsystem throughput that can be yielded for mixed traffic is 46.4% of the channelcapacity.6-13

Chapter 6Optimisation of CRA using adaptive CSAs pWith reference to Figure 6.7, we can appreciate that with an offered load over 46% ofthe channel capacity (produced by 34 stations), the highest system throughput wasranged from ≈ 45.5 to 46% with the configurations Forced-CSA(FSs-2, CSs-2) andForced-CSA(FSs-2, CSs-3). The maximum deviation between simulation and theoreticalresults was under 1% of the cc with an offered load above 46%.46.546.045.5System Throughput (%)45.044.544.043.543.042.542.041.5Simple-CSA(CSs-3)Simple-CSA(CSs-5)Forced-CSA(FSs-2,CSs-2)Forced-CSA(FSs-2,CSs-3)Variable-CSA(CSs-2)Variable-CSA(CSs-3)3 Mbps Upstream32 Kbps kbps IP IP &9.7 Kbps kbps VoIPBackoff CRA31 32 33 34 35 36 37 38 39 40umber of Active StationsFigure 6.7 – System throughput vs. No. of active stations.Exponential backoff algorithm with varying CSA applied.6-14

Chapter 6Optimisation of CRA using adaptive CSAs pOn high traffic loads, with this enhanced CSA only the contention slots needed toresolve the collisions are sent, allocating on average from 2.821 to 2.783 per request(according to the current traffic load), as shown in Figure 6.8, optimising further thebandwidth to be allocated to the reservation and contention-based access regions. InAverage umber of ContentionSlots per Request4.54.3Simple-CSA (CSs_3)Simple-CSA (CSs_5)4.1Forced-CSA (FSs_2,CSs_2)3.9Forced-CSA (FSs_2,CSs_3)V ariable-CSA -(CSs_2)3.7V ariable-CSA -(CSs_3)3.53.33.13 M bps Ups tream32 kbps Kbps IP IP &&2.99.7 kbps Kbps VoIPBackoff CRA2.729 30 31 32 33 34 35 36 37 38 39 40umbe r of Active StationsFigure 6.8 – Average No. of CSs per reservation request vs. No. of active stations.Exponential backoff algorithm with varying CSA applied.Mean Access Delay (ms)1000900800700600500400300200Simple-CSA(CSs-3)Simple-CSA(CSs-5)Forced-CSA(FSs-2,CSs-2)Forced-CSA(FSs-2,CSs-3)Variable-CSA(CSs-2)Variable-CSA(CSs-3)3 Mbps Upstream32 Kbps kbps IP IP &9.7 Kbps kbps VoIPBackoff CRA100031 32 33 34 35umber of Active StationsFigure 6. 9 – Mean access delay vs. No. of active stations.Exponential backoff algorithm with varying CSA applied.6-15

Chapter 6Optimisation of CRA using adaptive CSAs pCumulative Probability1.00.90.80.70.60.50.40.30.20.10.03 Mbps Upstream32 Kbps kbps IP &9.7 Kbps kbps VoIPBackoff CRA33 67 100 133 167 200 233 267 300 333 367 400 433 467 500Mean Access Delay (ms)terms of packet delays and number of streams supported, from Figures 6.9 and 6.10, thelowest access delays were gained using the Forced-CSA.For instance, with 45% of the channel capacity (produced by 33 stations), tolerablemean access delays for VoIP streams (under 50 ms) were seen only with Forced-CSA(FSs-2, CSs-3), supporting up to 33 stations. With a slightly increase in offered load(e.g. 46% produced by 34 stations), only the Simple-CSA (CSs-5) and the Forced-CSA(FSs-2, CSs-3) yielded tolerable low delays for the support of IP traffic.Approximately 73% of all data packets were transmitted in less than 200 ms asillustrated in Figure 6.10. The Simple-CSA (CSs-5) produced relatively low packetaccess delays because it sent more CSs than currently needed to resolve collisions,resulting in a reduction on the average CRGC. On the other hand, the Simple-CSA (CSs-5) wasted many CSs, trying to minimise the CRGC and therefore a slight decrease insystem throughput was obtained.Simple-CSA (CSs-3)Forced-CSA (FSs-1,CSs-3)Forced-CSA (FSs-2,CSs-3)Simple-CSA(CSs-5)Fo rced-CSA(FSs-2,CSs-2)Variable-CSA(CSs-2)Figure 6.10 – Cumulative probability vs. No. of active stations.Exponential backoff algorithm with varying CSA applied.Offered load = 46% produced by 34 stations.As a final remark for this section, in general the Forced-CSA outperforms the other twomechanisms (Simple and Variable-CSA). The Forced-CSA not only provided the lowestaccess delays, but also in most of the cases simulated provided the highest system6-16

Chapter 6Optimisation of CRA using adaptive CSAs pthroughput. This makes the Forced-CSA the optimal slot allocator for different trafficconfigurations.6.3 Dynamics of the DVB Contention Resolution AlgorithmsIn this section, we pay attention to the details of the exponential backoff algorithm andthe splitting tree algorithm adopted by the DVB/DAVIC communications protocol.6.3.1 Exponential backoff algorithm optimisationFor CATV networks that make use of the exponential backoff algorithm, theperformance of such networks is also determined by the initial and truncated backoffvalues defined (also referred to as backoff window: Bw[i-t]). Thus, the aim of thissection is to provide a performance analysis for different backoff windows and toindicate which windows offer the highest system performance. Two trafficconfigurations (mixed traffic and VoIP traffic) will be used.In order to achieve optimum system performance, the Forced-CSA (introduced inSection 6.2.3.1) will be utilised. With the exception of the minimum and maximumbackoff values, for this analysis we have used the same simulation parameters presentedin Table 5.4 (Chapter 5).6-17

Chapter 6Optimisation of CRA using adaptive CSAs pa) Results for mixed traffic using the exponential backoff algorithmThis analysis deals with mixed traffic at 41.7 kbps per station (32 kbps Internet plus 9.7kbps VoIP traffic). Figure 6.11 presents the cumulative probability plotted againstaccess delay close to saturation. This performance was recorded with 34 stations,producing an offered load of 46% of the channel capacity.From Figure 6.11, it can be seen that by defining short values for the initial backoffexponent (e.g. Bw[2-3] or Bw[2-5]) a poor system efficiency is obtained. Up to 33% ofall packets generated were transmitted in less than 100 ms. This is because thebacklogged NIUs are forced to transmit in any of the next 4 contention slots, therebyincreasing the risk of collision with new incoming packets.Similar inefficiencies are seen with large values for the initial and truncated backoffexponent parameters (e.g. Bw[5-7). The consequence of this being to cause backloggedNIUs to wait for a relatively long period before they can transmit a request, whichresults in many contention slots passing without being used.6-18

Chapter 6Optimisation of CRA using adaptive CSAs pBy defining intermediate values (e.g. Bw[4-6], Bw[3-4], B[3-5] and Bw[2-4]), a goodsystem performance can be obtained. However, the backoff window that offered bestsystem performance appeared to be Bw[4-6].a) Results for VoIP traffic using the exponential backoff algorithmThis second traffic configuration emulates pure VoIP traffic where a large network size(up to 110 active stations) is considered to cover saturation points. Each station nowproduces 9.7 kbps CBR traffic. The point of saturation for this traffic configurationvaried according to the backoff windows selected. Therefore, the analysis is presented interms of mean access delays and system throughput.We can see in Figure 6.12 that the network load becomes busy at ≈ 31% of cc (producedby 100 active stations). At this point, all backoff windows offered mean access delaysunder 20ms. However, with a higher offered load of ≈ 33% (yielded by 106 stations) thebackoff window that still offered mean packet access delays of under 50 ms withmaximum system throughput was Bw[5-7]. The backoff window Bw[4-8] also offered aCummulative ProbabilityMean Access Delay (ms)1.00.90.80.75000.64500.54000.4350Bw[4-6]Bw[2-4]Bw[3-4]Bw[3-6]0.3300Bw[4-6] 3 Mbps Upstream0.225032 Kbps kbps IP &9.7 Kbps kbps VoIP0.12003 Mbps UpstreamBackoff CRA0.01509.7 Kbps kbps VoIPBw[4-8]100 33 67 Backoff 100CRA133 167 200 233 267 300 333 367 400 433 467 500500Bw[2-5]Bw[2-4]Bw[3-4]Bw[3-5]Access Delay (ms)Bw[3-6]Bw[5-7]Bw[2-5]Bw[5-7]Bw[3-5]B w[2-3]Figure 6.11 – Cumulative probability vs. No. of active stations for different backoffwindows and offered load of 46% and mixed traffic.umber of Active StationsFigure 6.12 – Mean access delay vs. No. of active stationsfor different backoff windows and VoIP traffic.97 100 103 106 1096-19

Chapter 6Optimisation of CRA using adaptive CSAs pgood performance. In general, simulation showed us that when there are more than 70active stations, a good performance is achieved if we set the initial backoff value in therange [4-5] and the truncated backoff value in the range [6-8].From the previous analysis (for mixed traffic), the maximum system throughputsupported by the network has been reduced from 45% down to 32% of the channelcapacity (see Figure 6.13). This is due to the extended number of stations supported,which produce a larger number of collisions when competing for contention-based slots,and also because the mean packet size has been decreased from 368 bytes, (which is theormalised System Throughput0.353Mbps Upstream0.33 9.6 9.7 Kbps kbps VoIP V Backof f CRAB w[ 4 - 8 ]0.31B w[ 5 - 7 ]B w[ 3 - 6 ]0.29B w[ 4 - 6 ]0.270.25B w[ 2 - 5 ]B w[ 3 - 5 ]0.230.210.19B w[ 2 - 4 ]0.17B w[ 3 - 4 ]0.1591 94 97 100 103 106 109umbe r of Active StationsFigure 6.13 – System throughput vs. No. of active stationsfor different backoff windows and VoIP traffic.average packet size used for Internet traffic, as indicated in Section 5.3.1), to 146 byes(used for VoIP traffic), leading to a higher protocol overhead and a reduction in systemthroughput.6.3.2 Splitting tree algorithm optimisationIn CATV networks, INAs and NIUs that support a splitting tree algorithm, achieve abetter system performance by reducing the collision risk when transmitting reservationrequest messages at the cost of a higher processing times and complexity. This increasein performance is obtained by i) dividing one upstream contention-based slot into threeindependent minislots that carry shortened request messages and ii) providing two6-20

Chapter 6Optimisation of CRA using adaptive CSAs pregions of contention-based slots: the first region is normally used to resolve collisionsof backlogged stations, where NIUs with new incoming packets are not allowed totransmit reservation requests in these contention-based slots; and the second region isused only to transmit requests of new incoming packets.When selecting the splitting tree algorithm, a factor that contributes also to theperformance of the network is the ‘Entry Spreading’ (Es) value, introduced in Section3.5.3. This parameter can be either computed or fixed at the INA. For general purposes,the value of the Es factor will be fixed and ranged from 4 to 9 in our simulations. Weuse here the same traffic types as used for the previous algorithm (Internet and VoIPtraffic). The simulation parameters presented in Table 5.4 were also utilised for thisanalysis with the exception of the Es, which is the variable for this part of thediscussion.a) Results for mixed traffic using the splitting tree algorithmThe performance impact in terms of mean access delay, when varying the Es, ispresented in Figure 6.14. We can observe that for Es-8 and Es-6 tolerable delays under50 ms (for VoIP) are yielded when the offered load reaches 54% of the cc (generatedby 40 stations). By increasing the offered load by just two stations (to 42 NIUs, 57% ofMean Access Delay500450400350300250200150100500Es-43 Mbps Upstream32 kbps Kbps IP IP &9.7 Kbps kbps VoIPTree CRA38 40 42 44umber of Active StationsEs-8Es-7Es-6Figure 6.14 – Mean access delay vs. No of active stationsfor different Entry spreading factors and mixed traffic.6-21

Chapter 6Optimisation of CRA using adaptive CSAs pthe cc), very large delays are evident. At this point, the Es value that still offersacceptable delays (for IP traffic only) under 500 ms was Es-6. The other two values Es-5 and Es-9, (not included in Figure 6.14), showed the same results as for Es-7 and Es-8,respectively.In terms of system throughput, the difference of using distinct Es values was much lesssignificant. In general, regardless of the Entry spreading factor, the maximum systemthroughput ranged between 54.5 and 55% of the cc.b) Results for VoIP traffic using the splitting tree algorithmResults from simulations for VoIP traffic (see Figure 6.15), revealed that delays ofunder 30 ms can be obtained for all values of the Es factor, given an offered load up to46% of the cc (with 145 active stations). Optimum system performance is obtained withEs-6 and Es-7, achieving a maximum system throughput over 45% of the cc, as shownin Figure 6.16. In general, it was found that by defining small values for the Entryspreading factor (e.g.: Es-4), at high traffic loads (above 45% of the cc), NIUs with newarrivals are forced to transmit in one of the next three available contention-basedminislots, reserved for new incoming packets with probability of 3/4 = 0.75. This leadsto a higher risk of collision when more than two NIUs are competing for requesttransmission.10009008003 Mbps Upstream9.7 Kbps kbps VoIPTree CRAMean Access Delay700600500400300200Es-4Es-6 / Es-7Es-8/ Es-91000140 145 150 155umber of Active StationsFigure 6.15 – Mean access delay vs. No. of active stationsfor different Entry spreading factors and VoIP traffic.6-22

Chapter 6Optimisation of CRA using adaptive CSAs pSystem Throughput (%)45.445.3Es-6 / Es-745.245.1Es-8/ Es-945.044.944.8Es-444.744.644.53 Mbps Upstream44.49.7 Kbps kbps VoIP44.3Tree CRA44.2145 150 155 160 165 170umber of Active StationsFigure 6.16 – System throughput vs. No. of active stationsfor different entry spreading factors and VoIP traffic.By defining larger values for the Entry spreading factor, (e.g. Es-8 and Es-9), theprobability of transmission of requests in one of the next 3 contention minislots is (3/8≈) 0.38 and (3/9 ≈) 0.33 for Es-8 and Es-9, respectively. This is equivalent to passing(on average) 2 or 3 contention slots before transmitting a request, which also results in aslight reduction in system performance.Results reported in this section suggest that for optimum performance, the Es factorshould be set to 6 for mixed traffic (or medium size networks) and 6 or 7 for VoIPtraffic (or large networks).6.3.3 Performance comparison between the exponential backoffalgorithm and the splitting tree algorithmIn the previous analysis of Chapter 5, we presented an initial performance comparisonbetween both CRAs of the DVB/DAVIC protocol. In this section we focus explicitlyon the details and dynamics of each CRA in order to demonstrate why the splitting treealgorithm outperforms the exponential backoff algorithm.6-23

Chapter 6Optimisation of CRA using adaptive CSAs pResults presented in Figures 6.17 to 6.20 show a performance comparison between theexponential backoff algorithm and the splitting tree algorithm, for three different trafficconfigurations (32 kbps IP, 9.7 kbps VoIP and 41.7 kbps mixed traffic).Figure 6.17 indicates that the increase in system throughput when using the splittingtree algorithm is 9%, 12.3% and 9.7% of the cc for Internet, VoIP and mixed traffic,respectively. Figure 6.17 also shows that the maximum throughput achieved varies from32% to 58% of the cc.System Throughput (%)6560555045403530BackoffTreeVoIP TrafficI PTrafficMixedTraffic2525 28 31 34 37 40 43 46 49 52 55 58 61 64Offered Load (%)Figure 6.17 – System throughput. Exponential backoff vs. splitting tree algorithmfor different traffic configurations.6-24

Chapter 6Optimisation of CRA using adaptive CSAs pFrom Figure 6.18, by taking the limits before the network became unstable, the increasein terms of supported streams corresponded to 8, 39 and 6 stations for Internet, VoIPand mixed traffic, respectively.In terms of efficiency, results not shown in this analysis indicate that the maximumchannel utilisation varied from 76% to 86% of the cc for the exponential backoffalgorithm and from 88% to 90% of the cc for the splitting tree algorithm. Theexponential backoff algorithm achieved a lower channel utilisation due to theconsiderable number of unused forced-contention slots that were allocated to resolvecollisions, especially when a large network size was considered (as it was the case ofVoIP traffic).Another important comparison is that according to the Contention-Resolution-GrantCycle introduced in Section 3.5.1, there are three types of delays involved in thetransmission of data packets, such as waiting delay, contention delay and grant delay, asdepicted in Figure 3.3. The exponential backoff algorithm yields a low systemperformance mainly because this algorithm requires more time to resolve collisionMean Access Delay (ms)500450400350300250200150100500BackoffTreeVoIP TrafficMixedTrafficI PTraffic25 28 31 34 37 40 43 46 49 52 55 58Offered Load (%)Figure 6.18 – Mean access delay. Exponential backoff vs. splitting tree algorithmfor different traffic configurations.(referred to as contention delay) than its counterpart splitting tree algorithm.6-25

Chapter 6Optimisation of CRA using adaptive CSAs pIn order to make a direct comparison among the components of delays involved in thetransmission of data packets, we can analyse the situation when only IP traffic isdelivered. From Figure 6.19, on high traffic loads, the splitting tree algorithm requireson average 10 ms to transmit successfully a reservation request (this corresponds to themean contention delay curve in Figure 6.19b), in comparison to approximately 19 msrequired by the exponential backoff algorithm (Figure 6.19a.). This variation of 9 msmakes a difference in system performance, because having a large service time(comprised of the mean contention delay plus the mean grant delay) results in packetsbeing queued for a longer period of time as the offered load increases further, which inturn results in large mean access delays.Components of Delay (ms)1201101009080706050403020100a) Exponential Backoff Algorithm b) Splitting Tree AlgorithmMeanwaitingdelayMeangrantdelayMeancontentiondelay42 43 44 45 46 47 48 49 50 51 52umber of Active StationsMeanwaitingdelay48 49 50 51 52 53 54 55 56 57 58 59 60umber of Active StationsMeangrantdelayMeancontentiondelayFigure 6.19 – Components of delay, Exponential backoff vs. splitting tree algorithmfor IP traffic.6-26

Chapter 6Optimisation of CRA using adaptive CSAs pThis effect can be appreciated in Figure 6.20, which presents in detail the collision ofthree NIUs. For this analysis the period begin when three stations collide and ends whenthe NIUs transmit their data message successfully.In Figure 6.20 each line shows the use of the 18 slots described in each signalling frame(MCI). For the exponential backoff algorithm, stations 43, 25 and 8 transmit areservation request at MCI 1, using the same contention slot. Then, the stations shouldwait until the headend sends back the acknowledgements of these reservation requests,which is at MCI 3. At this point, the stations detect the collision. Station 43 backs offfor 10 CSs, and stations 8 and 25 transmit in CS 0 and 1, respectively.3a) a) Exponential Backoff Algorithmb) Splitting Spliting Tree Algorithmtime (ms) 691215 18212427⌨30333639 42 45 4860Successful requestslot/minislotCollidedslot/minislotEmptyslot/minislotRersed Reserved data data slot slotFigure 6.20 – Upstream slot usage. Exponential backoff vs. splitting tree algorithmfor IP traffic with 49 stations. Sequence captured at 30 seconds of simulation.6-27

Chapter 6Optimisation of CRA using adaptive CSAs pStation 25 transmits successfully and then waits until the headend grants its request,which is at MCI 11. This is because the headend, at this point, was serving requestsfrom other stations not shown. Station 8 collides two more times at MCI 3 (with station1) and MCI 8 (with station 11). In MCI 16, station 8 transmits a successful request,which is granted in MCI 20. Station 43 transmits successfully at MCI 4 and its requestis served in MCI 13. This algorithm required 16 signalling frames (48 ms) to resolve acollision among three stations.Lets us now analyse the case when the splitting tree algorithm is utilised. At MCI 1,there are also three stations that collide (stations 28, 11 and 4). At MCI 3, the headendallocates one more CS (in the first position of the MCI), which is used only among thecollided stations. So, all stations transmit again after two MCIs once the collision hasbeen detected. At this moment, station 4 transmits a successful request using the firstminislot of this additional CS (scheduled for MCI 12). Station 11 and 28 collide againusing the third minislot. It is at MCI 5 when they transmit successfully and servedcompletely at MCI 15.Generally, the splitting tree algorithm is much more efficient because when a collisionhappens (e.g. between two or even three stations), it takes on average from 2 to 5 MCIcycles to resolve, whilst the exponential backoff algorithm takes from 2 to 16 (and sometimes longer). Another advantage of the splitting tree algorithm is that it requires fewercontention slots than the exponential backoff algorithm in order to transmit successfulrequests. From Figure 6.20, we can appreciate that the number of contention slots usedby the splitting tree algorithm ranged from 2 to 3 CSs, in comparison to 3 to 7 CSsrequired by the exponential backoff algorithm in each MCI cycle.The same performance analysis and a comparison between these two contentionresolutions algorithms, for an upstream channel of 6.176 Mbps, has been reported in[86] and summarised in Figure E.1 of Appendix E.6-28

Chapter 6Optimisation of CRA using adaptive CSAs p6.4 ConclusionsIn this chapter we have shown that the overall system performance of the DVB/DAVICcable communications system can be significantly improved by the use of three novelContention Slot Allocators (Simple-CSA, Variable-CSA and Forced-CSA), whichallocate dynamically the number of contention slots that should be allocated in the nextsignalling frame, based on the current traffic load. Results presented in this chapter havepointed out that the Forced-SCA not only provides the highest system throughput but inmost of the cases also offers the lowest packet access delays for the exponential backoffalgorithm. Simulation results have been shown to be accurate when compared withresults from theoretical analysis.In addition, a performance comparison of the contention resolution algorithms adoptedby the DVB/DAVIC protocol was also presented. The splitting tree algorithm takesadvantage of the exponential backoff algorithm in the sense that feedback and allocationinformation allows a station, (with new incoming arrivals) to compete for contentionbasedslots without the risk of collision with backlogged stations. One more advantageof the splitting tree algorithm is that the use of minislots for reservation requests furtherdecreases the risk of collisions, since one contention-based slot is divided into threeminislots, increasing the probability of successful request transmissions. However,drawbacks of the splitting tree algorithm are higher complexity at the headend,increased processing times of the feedback and allocation information at the station, andsince every contention slot should be acknowledged regardless of whether it is used ornot, higher control information at the downstream channel is assigned.Results presented by both contention resolution algorithms showed that the systemperformance is a trade-off between mean access delay/system throughput and the valuesselected for the initial/truncated backoff values and the Entry Spreading factor.In general, results presented here showed that an increase of over 9% on systemperformance can be obtained by the use of a splitting tree algorithm when backoffvalues (initial/truncated) and the Entry spreading factor have been optimised, fordifferent traffic configurations such as Internet, VoIP and mixed traffic.6-29

Chapter 7PERFORMANCE OPTIMISATION FORTHE SUPPORT OF TCIS AND APERFORMANCE COMPARISON OFDVB/DAVIC AND DOCSIS7.1 IntroductionThe DVB/DAVIC protocol specification supports only a limited reservation accessmechanism (referred to as ‘pure reservation access’ - PRA) and has not yet beenoptimised for the delivery of isochronous streams. The functionality of the PRA waspreviously touched on under the CRGC, introduced in Section 3.5.1.Some contention resolution algorithms, such as the exponential backoff algorithm,produce relatively high access delays for upstream transmissions. The splitting treealgorithm, which uses less contention slots and shortened reservation requests, rapidlyresolves the contention resolution cycle, thus reducing access delays and improvingsystem throughput. However, the splitting tree algorithm combined with optimisedconfiguration parameters as discussed in Chapters 5 and 6 is not ideal for the support ofthe next generation timing critical interactive services.Therefore, in this chapter some novel improvements and techniques are introduced,which will enable the DVB/DAVIC MAC protocol to provide the delay requirementsoptimally for the delivery of delay sensitive services. The objective is to achieve anincreased performance by introducing new reservation request techniques that reduce oravoid the increased risk of collisions during congestion periods.A second approach involves QoS with a guaranteed delivery. Although, this techniqueis only supported by the second version of EuroModems.7-1

Chapter 7Performance optimisation for the support of TCIS pFinally, a performance comparison between the two standardised cable communicationssystems, DVB/DAVIC and DOCSIS, is also considered. Here, the major characteristicsand the fundamental performance properties of these two leading protocols are thusevaluated.7.2 Enhanced-Reservation-Request MechanismsSix improved mechanisms are now discussed. Each mechanism reduces access delaysand increases overall system performance required in support of TCIS and also highspeedbursty traffic. All six use only the reservation access mode of the DVB/DAVICprotocol.7.2.1 Reserved Request (RR)The ‘Reserved Request (RR)’ mechanism follows the same principle as the purereservation access request. A difference is that the RR allows a station to request one slotmore than currently needed in order to transmit a further request as depicted inFigure 7.1. This additional slot is only requested when there are other data packets inthe transmission queue. Thus, the NIU can indicate its request for additional bandwidthwithout any collision risk between two data messages. A risk of collision exists whenthe transmission queue is empty and a new data message arrives.Message 3(3-ATM cells)Message 2(5-ATM cells)Message 1(2-ATM cells)PRA: Reservation Request = 2 SlotsRR: Reservation Request = 3 SlotsPRA: Reservation Request = 5 SlotsRR: Reservation Request = 5 SlotsFigure 7.1 – Functionality of the RR mechanism for different packet sizes in thetransmission queue.7-2

Chapter 7Performance optimisation for the support of TCIS p7.2.2 Continuous Reserved Request (CRR)The ‘Continuous Reserved Request (CRR)’ mechanism allows a station to request onemore slot than needed, regardless of the state of the transmission queue (see Figure 7.2).If the transmission queue of the NIU is empty, then the extra slot is not used. Thismechanism may reduce the mean access delays, because for every two MCI cycles (6ms), there is a reserved slot in which to place a request. A drawback of this mechanismis that unused extra slots cause a considerable waste of bandwidth during congestionperiods.RR: Reservation Request = 3 SlotsRR: Reservation Request = 5 SlotsCRR: Reservation Request = 3 SlotsCRR: Reservation Request = 6 SlotsFigure 7.2 – Functionality of the CRR mechanism for different packet sizes in thetransmission queue.7.2.3 Enhanced Pure Reservation Access (EPRA)A disadvantage of the pure reservation access, the RR and the CRR requestmechanisms, is that they are all based on a per-packet transmission basis rather than aper-block transmission. In other words, a station only requests the number of slotsneeded to transmit the current data message (up to 256 ATM cells). As the offered loadis increased, short messages tend to accumulate and then saturate the NIU buffer. Thus,long packet access delays render the system unstable. This effect is true even at lowerloads but does depend on of the traffic type.7-3

Chapter 7Performance optimisation for the support of TCIS pThe ‘Enhanced Pure Reservation Access (EPRA)’ mechanism has been designed to supportper-block transmissions and to allow a station to request slots of up to 256-ATM cells, so thatmost of the packets waiting in queue can be sent using a single reservation request. This isshown in Figure 7.3.PRA: Reservation Request = 2 SlotsEnhanced PRA: Reservation Request = 10 SlotsFigure 7.3 – Functionality of the EPRA mechanism for different packet sizes in thetransmission queue.7.2.4 Piggyback Request (PG)The ‘Piggyback Request (PG)’ mechanism avoids reservation requests between datamessages. When there are more data packets in the transmission queue waiting to betransmitted, two unused bits of the ATM header (in the upstream slots) are used to carrythe slot request. This is illustrated in Figure 7.4.Data SlotUWATMHeaderInform ation P ayloadRSG Bunused bitsbits8 7 6 5 4 3 2 1G FC VPIVPI +VCIPT CLPH E CPiggybackRequest00 -01 -10 -11 -Pg V alue 1 (0-slots )Pg V alue 2 (e.g. 4-slots)Pg V alue 3 (e.g. 8-slots)Pg V alue 4 (e.g. 12-slots)Figure 7.4 – Data slot structure with a piggyback request.7-4

Chapter 7Performance optimisation for the support of TCIS pFour states are defined. These are mapped to indicate four distinct numbers of requestedslots, for example 0, 4, 8 and 12, rather than the actual number of slots required [101].Figure 7.5 shows the mechanism regime.The PG mechanism (which is also supported by the DOCSIS protocol) tends to be veryefficient for the delivery of bursty traffic [120].PG(0,4,8,12): Piggybacked Request = 12 Slotsplaced in the last ATM cell of the first messagetransmittedPG(0,4,8,12): Piggybacked Request = 0 Slotsplaced in the last ATM cell of the first messagetransmittedFigure 7.5 – Functionality of the PG mechanism for different packet sizes in thetransmission queue.7.2.5 Continuous Piggyback Request (CPG)The ‘Continuous Piggyback Request (CPG)’mechanism allows a station to alwayspiggyback a request and prevents the station from falling back into the contentionrequest mode. This is shown in Figure 7.6. At least one slot is requested regardless ofthe state of the transmission queue. The CPG mechanism is efficient but not ideal forCBR traffic. This is because there are only four possible values for the piggybackrequest. Unused slots cause wasted bandwidth and particularly when a large number ofNIUs are active.CPG(1,4,8,12): Piggybacked Request = 12 SlotsCPG(1,4,8,12): Piggybacked Request = 1 SlotsFigure 7.6 – Functionality of the CPG mechanism for different packet sizes in thetransmission queue.7-5

Chapter 7Performance optimisation for the support of TCIS pIn addition, during long idle periods, the continuous piggyback mechanism fails due tothe excessive number of slots that remain unused. A timer function stops thismechanism after a time threshold is exceeded. Time periods of up to 2.286s can bebridged without using contention slots [35].7.2.6 Unsolicited Grant Slot (UGS)The ‘Unsolicited Grant Slot (UGS)’ mechanism provides a station with a fixed numberof slots at periodic intervals, using the reservation grant message. The stations must useonly unsolicited grant slots for upstream transmissions. The number of slots needed by astation and the periodic intervals can be negotiated during the connection set-up. Thismechanism can be used only when the CBR parameters are satisfied regarding theactual data rate, packet size and packet interval. In addition, the UGS mechanismcompletely avoids the transmission of reservation requests and allocates morebandwidth to the reservation-based region, thereby reducing significantly the packetaccess delay and increasing the overall system throughput.The continuous reserved request, piggyback request and continuous piggyback requestmechanisms were first reported in [43], where a comprehensive performance analysis interms of the cumulative probability of packet delay was presented. Due to theirenhanced performance these three mechanisms have now been integrated into the latestversion of the DVB/DAVIC standard ‘ETSI EN 200 800’ [35].The unsolicited grant slot, enhanced pure reservation access and reservation requestmechanisms are novel and have been developed in this project. Results derived fromthis research show that the UGS mechanism significantly outperforms the othermechanisms in terms of system performance and can be easily adopted in the ETSIEN 200 800 protocol specification.7-6

Chapter 7Performance optimisation for the support of TCIS p7.2.7 Performance comparison of Enhanced-Reservation-RequestMechanismsTwo different traffic situations are presented in order to give an overview of theperformance of the enhanced-reservation-request mechanisms in reservation-accessmode. For this analysis both contention resolution algorithms were used.The traffic types were as previously seen and consisted of VoIP traffic at 9.7 kbps andmixed traffic at 41.7 kbps. Table 7.1 shows the values of the simulation parameters thatwere shown to have provided optimum system performance in Chapters 5 and 6.Table 7.1 – Optimised simulation parameters.Simulation ParameterValueUpstream data rate (QPSK)3.088 MbpsDownstream data rate (64-QAM, In band)42 MbpsBuffer capacity per NIU3000 ATM cellsHeadend and NIU processing delay2 microseconds eachMaximum contention access message length0 ATM cellsMaximum number of active NIUs (EuroModems) 180Maximum reservation/piggyback32 for PRA, EPRA, RR, CRR, and 256 for EPRArequest length (in slots)0,3,6,12 for PG and 1,3,6,12 for CPGBackoff windows (exponential backoff algorithm)[4-6] for Mixed traffic and [5-7] for VoIP trafficMinimum CSs per signalling frame (exponential backoff algorithm) 3 for PRA, EPRA, RR, PG and 1 for CRR, CPG, UGSContention slot allocator used (exponential backoff algorithm)ForcedEntry spreading factor (splitting tree algorithm) 6Minimum CSs per signalling frame (splitting tree algorithm)2 for PRA, EPRA, RR, PG and 1 for CRR, CPG, UGSPropagation delay (coax and fibre)5 µs/KmSimulation time for each run60sDistance from nearest/farthest NIU to the Headend10-16 Km, randomly distributedTransmission time (cycle) of the signalling frame3 ms7-7

Chapter 7Performance optimisation for the support of TCIS pA) Performance analysis for the exponential backoff algorithmFrom simulation results, Figures 7.7 to 7.9 present a performance comparison betweenthe enhanced-reservation-request mechanisms when the exponential backoff algorithmis used.Results for mean access delay showed that on high traffic loads, above ≈ 33% of the cc(105 stations) for VoIP traffic (Figure 7.7a) and 46% of the cc (34 stations) for mixedtraffic (Figure 7.7b), all the enhanced mechanisms offered lower access delays than thepure reservation access mode. For both traffic types, it was found that the bestmechanism is the continuous piggyback request. This mechanism achieved a betterperformance because all requests are piggybacked on the same data messages. Also, ifthe buffer is empty when a station receives the piggybacked slots, such reserved slotsare then used to send further requests for at least one slot. Contention access is thereforeavoided. Thus, more bandwidth can be allocated to the reservation access region. Thisresults in the lowest access delays (during high traffic loads) and the highest systemthroughput.Mean Access Delay (ms)500450400350300250200150100503 Mbps Up.9.7 Kbps kbps VoIPBackoff CRAa) VoIP Traffic b) Mixed TrafficPRARRCRREPRAPGCPGPRARRCRRPG3 Mbps Up.Mixed TrafficBackoff CRAEPRACPG070 80 90 100 110 120 130 140 150 160 170umber of Active Stations31 33 35 37 39 41 43 45umber of Active StationsFigure 7.7 – Mean access delay vs. No. of active stations for VoIP and mixed traffic.Exponential backoff algorithm with enhanced-reservation-request mechanisms.7-8

Chapter 7Performance optimisation for the support of TCIS pFor mixed traffic on loads of under 46% of the cc (34 stations), the CPG mechanismalso provided the lowest access delays. However, for the transmission of VoIP streamson loads of under 33% cc (105 stations), it did not offer the lowest access delaysbecause it was seen that with a large number of active stations, the headend has to granttwo types of requests. These are the requests made for one slot, which are then used tosend a further reservation request and the requests made for the transmission of datamessages. In simulation, it was seen that for low traffic loads, most requests are madefor one slot, which delay the other requests for data messages. Thus, the NIUs have towait longer in order to transmit their data slots.A similar effect is also appreciated with CRR. However, in this case the number ofrequests made for one slot is larger than with the CPG mechanism. This is because thereservation requests are made for the transmission of the current data packet, and theCPG not only requests bandwidth for the current data message but also for other packetsin the queue, as long as the number of slots requested does not exceed the maximumpiggybackedrequest threshold (set to 12 slots). Hence, mean access delays are slightlylarger than those produced by the CPG and the other mechanisms.In terms of number of VoIP streams supported, with the PRA and PG mechanisms only105 connections can be upheld with mean access delays of under 50 ms. That numbercan be increased to 110 connections with the use of the RR and EPRA mechanisms.However, with the CPG mechanism the network capacity can be extended up to 124connections.Results for mixed traffic indicate that the number of VoIP streams supported in thepresence of Internet traffic is 32 connections with the PRA mechanism. Simulationshave also shown that this number can be extended by up to 8 connections using theenhanced-reservation-request mechanisms.7-9

Chapter 7Performance optimisation for the support of TCIS pIn terms of throughput, Figure 7.8 shows the maximum system throughput achieved byeach mechanism. The PRA and RR achieve under ≈38% cc for VoIP traffic(Figure 7.8a) and under 47% of the cc for mixed traffic (Figure 7.8b). This is due tocontinuous contention access. The maximum achievable throughput is produced by theCPG mechanisms, which can be up to 52% and 61% of the channel capacity for VoIPand mixed traffic, respectively.All of the enhanced mechanisms achieve a higher system throughput than the PRAmechanism, lowering collision risk and reducing (or even avoiding) contention accesstransmissions, thereby allocating more bandwidth to the reservation access region.System Throughput (%)565350474441383532292623203 Mbps Upstream9.7 Kbps kbps VoIPBackoff CRAa) Voice TrafficCRRRRPRACPGPGEPRA70 85 100 115 130 145 160 175umber of Active Stations6260585654525048464442403 Mbps UsptreamMixed TrafficBackoff CRAb) Mixed TrafficCPGPG & EPRAPRACRR31 33 35 37 39 41 43 45 47umber of Active StationsRRFigure 7.8 – System throughput vs. No. of active stations for VoIP and mixed traffic.Exponential backoff algorithm with enhanced-reservation-request mechanisms.7-10

Chapter 7Performance optimisation for the support of TCIS pIn Figure 7.9, we can appreciate how the amount of bandwidth wasted by collisions isdecreased considerably by the enhanced-reservation-request mechanisms as the offeredtraffic load becomes more intense. The continuous request mechanisms, (CRR andCPG) were designed to avoid contention access entirely. The RR and PG follow thesame dynamics as the CRR and CPG mechanisms respectively, with the exception thatwhen there are no packets in the buffer, stations fall back to contention access.Bandwidth Consumed by CollidedContention Slots (%)3027242118151296303 Mbps Up.9.7 Kbps kbps VoIPBackoff CRAa) Voice TrafficPRACRR & CPGEPRA70 85 100 115 130 145 160 175umber of Active StationsPGRR876543210CRR & CPGb) Mixed TrafficRREPRAPGPRA3Mbps Up.Mixed TrafficBackoff CRA31 33 35 37 39 41 43 45umber of Active StationsFigure 7.9 – Bandwidth consumed by collisions vs. No. of active stations for VoIP andmixed traffic. Exponential backoff algorithm with enhanced-reservation-requestmechanisms.7-11

Chapter 7Performance optimisation for the support of TCIS pB) Performance analysis for the splitting tree algorithmSimulations with the splitting tree algorithm, Figure 7.10a indicates that for pure VoIPtraffic optimum system performance may be obtained using the unsolicited grantmechanism. With this mechanism the network capacity is considerably increased to thehighest figure for this research at ≈ 170 connections. This equates to approximately 53%of the available channel capacity and produces mean access delays below 7 ms. Hence,this mechanism has been shown to match the requirements of CBR. The othermechanisms, with the exception of CPG, support the same number of connections equalto ≈ 145.Results for mixed traffic are shown in Figure 7.10b. It should be noted that discussionof the UGS mechanism is absent here since in a mixed traffic situation the constant bitrate requirements cannot be satisfied using this technique. From the figure we can alsoappreciate that the enhanced-reservation-request mechanisms are more effective withhigh traffic loads. For example, for traffic loads above 55% of the cc, the maximumnumber of Internet connections that can be supported was ≈ 45 with the CPGmechanism, compared with ≈ 41 with the PRA mechanism. For delay sensitiveapplications the number of connections supported resulted in ≈ 40, regardless of themechanism utilised.Mean Access Delay (ms)30027525022520017515012510075502503 Mbps Uptream9.6 9.7 Kbps kbps VoIPSpliting Splitting Tree CRAa) VoIP Traffic b) Mixed TrafficCPGEPRAPRA70 80 90 100 110 120 130 140 150 160 170umber of Active StationsPGUGS7-123 Mbps UpteamMixed TrafficSplitting Spliting Tree CRAPRAEPRACPG37 38 39 40 41 42 43 44 45 46 47umber of Active StationsFigure 7.10 – Bandwidth consumed by collisions vs. No. of active stations for VoIP andmixed traffic. Splitting tree algorithm with enhanced-reservation-request mechanisms.PG

Chapter 7Performance optimisation for the support of TCIS pIn general, the UGS and also the CPG (for small networks) are quite efficient and bestsuited for the delivery of VoIP streams. For Internet traffic, it was found that the CPG,PG and EPRA mechanisms, combined with the use of the splitting tree algorithm, aremore appropriate for the transmission of Internet and bursty traffic.With all of the enhanced-reservation-request mechanisms, it has been demonstrated thatthe number of VoIP streams and Internet connections can be significantly extended,using only the reservation access mode of the DVB/DAVIC communications protocol.This access scheme is the basic mode for the first version of EuroModems [36]).A drawback of using the reservation access mode for the provision of delay sensitiveapplications is that bandwidth cannot be guaranteed. This disadvantage is because alltraffic transmitted using this mode is treated with a ‘best-effort’ service and in somesituations, the delivery of TCIS requires a special treatment for an improved service.Therefore, in order to provide guaranteed bandwidth for TCIS, the use of QoS is crucialand is addressed in the following section.7.3 Quality of Service (QoS)This section outlines some issues related to QoS and how a cable network compliant tothe DVD/DAVIC protocol specification can make use of QoS characteristics for theprovision of guaranteed bandwidth. Here a comprehensive performance analysis forQoS is presented for the support of TCIS.In the simplest sense according to [82], QoS is the ability of a network element (e.g. anapplication, host or router) to have some level of assurance that service requirementscan be satisfied, providing a consistent predictable data delivery service.The Internet Protocol provides what is called a ‘best-effort’ service, making noguarantees about when data will arrive, or how much traffic can be delivered. Timingcritical interactive applications, including voice, audio and video streaming, demandhigh data throughput with low-latency in two-way communications.7-13

Chapter 7Performance optimisation for the support of TCIS pIn order to provide service guarantees, a level of quantifiably reliable service may needto be supplemented with the ability to differentiate traffic and enable different servicelevels. Two types of QoS are proposed. They are complementary and designed for usein combination in different network contexts. These schemes are known as‘prioritisation’ and ‘reservation’. The analysis presented in the following twosubsections examines to what extent performance can be improved upon whenprioritisation and reservation mechanisms are used.7.3.1 Prioritisation in QoSOne approach to QoS is to use the Type of Service (ToS)-based relative priorities of theIPv4 header (or the 4-bit Priority field of the IPv6 header), which indicates in a simpleway the relative delay and drop sensitivity of a packet, as depicted in Figure 7.11. Thismethod gives TCIS streams higher priority than data packets, but does not provide aguarantee of bandwidth or latency.Although traffic prioritisation is not part of the DVB/DAVIC specification, a fastertransmission for the delivery of TCIS streams can be provided by mapping the ToS fieldwith 2-levels of priority at the DVB/DAVIC MAC layer, as explained below.The ToS indicates the desired usage of the packet. The field itself contains a 3-bitprecedence indicator for the priority of the packet and 3 flags (D, T and R) to showwhether delay, throughput or reliability are relevant for the transmission. Most routersin the Internet are as yet set to ignore the ToS field.0 3 7 15 31Ver IHL Type of service Total lengthIdentification Flags Fragment offsetTime to live ProtocolSource addressHeader checksumDestination addressOption + PaddingDataPrecedenceType of service3-bits 1-bits 1-bits 1-bits 2-bits111 Network control110 Internetwork control101 CRITIC/ECP100 Flash override011 Flash010 Immediate001 Priority000 RoutineDelay0 Normal1 LowThroughput0 Normal1 HighReliability0 Normal1 HighReservedFigure 7.11 – IP header structure.7-14

Chapter 7Performance optimisation for the support of TCIS pFor DVB/DAVIC cable networks, this field can be used to provide a faster delivery ofTCIS streams by assigning a higher priority to such streams. In this way, a packet thatarrives at the NIU can be classified according to its ToS requirement and placed in aqueue with two or more levels of prioritisation.In this section, a mixed traffic situation with 41.7 kbps (as described in Section 5.3.3)was used in order to provide a performance analysis when prioritisation for QoS isconsidered. For this analysis only the splitting tree algorithm was used because of itssuperior performance over the exponential backoff algorithm. Results presented here arefor the pure reservation access and the piggyback mechanisms.From Figure 7.12a, it can be seen that packet access delays produced by the PRAmechanism when prioritisation is not supported, were almost the same for both traffictypes. The maximum number of stations supported was 40 for VoIP traffic (with meanaccess delays of under 5 0ms) and 42 for Internet traffic (with mean access delays ofunder 500 ms). This is to be expected, since both traffic types were treated equally.However, from Figure 7.12b when prioritisation was supported, on high traffic loads orcongestion periods (above 44 stations or 59% of the cc), it was seen that packet accessdelays for VoIP traffic produced by the PRA and the PG mechanisms remained atMean Access Delay (ms)3002752502252001751501251007550250a) Without Prioritisation - PRA b) With Prioritisation - PRA & PGInternetTrafficVoIPTraffic34 36 38 40 42 44 46 48 50 52 54umber of Active Stations7-15PRAInternetTrafficPGVoIP TrafficPRA34 36 38 40 42 44 46 48 50 52 54 56 58umbe r of Active StationsFigure 7.12 – Mean access delay vs. No. of active stations for VoIP and Internet traffic.Splitting tree algorithm with and without prioritisation.

Chapter 7Performance optimisation for the support of TCIS papproximately 50 ms, which is suitable for the support of audio streams with heavytraffic loads. Delays for Internet traffic were lower with the PG mechanism due to thecontinuous piggyback requests on high traffic loads, which avoids contention access.In terms of system throughput, when prioritisation is not used (see Figure 7.13a), wecan observe that the throughput for both traffic types increases almost linearly withrespect to the NIUs population up to the point of saturation (reported at 57% of the cc,with 42 stations) . After this, VoIP and Internet packets start accumulating in the NIU’squeues. Conversely, with prioritisation (Figure 7.13b), it can be seen that bothreservation request mechanisms transmit all VoIP streams in the first place and then useremaining bandwidth for Internet packets. As explained in Section 7.2.7, the PGmechanism achieves a higher system throughput than the PRA because of the reducednumber of collisions in periods of congestion.With very high traffic periods (above 73% of the cc or 54 stations), there is a slightdifference for VoIP traffic. With the PG mechanism after transmitting the VoIP streamssome piggybacks requests are made for Internet traffic. These requests at the headenddelay subsequent piggyback requests made for VoIP traffic. This is because the headendgrants all requests using a best-effort service. This problem can be solved by replacingSystem Throughput (%)50454035302520151050a) Without Prioritisation - PRA b) With Prioritisation - PRA & PGInternetTrafficVoIPTraffic34 36 38 40 42 44 46 48 50 52 54umber of Active StationsInternetTrafficVoIPTrafficPRAPGPGPRA34 36 38 40 42 44 46 48 50 52 54 56 58umbe r of Active StationsFigure 7.13 – System throughput vs. No. of active stations for VoIP and Internet traffic.Splitting tree algorithm with and without prioritisation.7-16

Chapter 7Performance optimisation for the support of TCIS pthe scheduling mechanism at the headend with one that supports at least two levels ofpiggyback or even reservation requests and also gives higher priority to VoIP traffic orTCIS applications in general.Although this prioritisation mechanism offers an improved performance for TCISapplications, a more efficient mechanism is examined in the following section.7.3.2 Reservation (Fixed-rate access) in QoSThe second approach for QoS with a guaranteed service, uses a fixed-rate accessconnection. In such connections data is sent in slots assigned at a fixed-rate basedaccess region in the upstream channel, as described in Section 3.5.4, Figure 3.6. Theseslots are uniquely assigned to a connection by the INA. The number of slots needed by aconnection and the periodic intervals are negotiated during the connection setup. Thecomplete set of procedures is illustrated in Figure 7.14.When an NIU requires a new fixed-rate connection or needs to change fixed-rateparameters, a Resource Request message is first sent to the INA by the NIU, includingthe new parameters. Example parameters are new cyclic assignment needed, requestedbandwidth, distance between slots and connection identifier (CID). The INA answerssuch requests by sending a Connect message to the NIU, indicating whether the newcyclic assignment is granted and if so, a new set of fixed-rate parameters are providedsuch as frame length, fixedrate start, fixedrate distance, fixedrate end and CID.Connection-SetupINA / HeadendResource RequestConnect MessageConnect ResponseNIU / EuroModemNew cyclic assignment needed(requested bandwidth, maximumdistance between slots and CID)Cyclic assignment granted(CID, frame length, fixedrate start,fixedrate distance and fixedrate end)CIDConnect ConfirmFigure 7.14 – Connection-setup for a Fixed-rate access connection.CID7-17

Chapter 7Performance optimisation for the support of TCIS pThe frame length parameter indicates the number of successive slots to use in the fixedrate access region associated with each fixed rate slot assignment. The fixedrate startparameter indicates the starting slot within the fixed rate access region that is assignedto the NIU. The NIU may use the next frame length slots of the fixed rate accessregions. The fixedrate distance parameter represents the distance between additionalslots assigned to the NIU and the fixedrate end parameter indicates the last slot that maybe used for fixed rate access.Subsequently, the NIU accepts the connection by sending a Connect Response message(indicating its CID) to the INA, which in turn answers this message by sending aConnect Confirmation message to the NIU.The analysis to follow will address the optimal system performance of the DVB/DAVICcable communications protocol when QoS (with a guaranteed service) is supported.Two traffic types were analysed: mixed traffic and VoIP traffic. For this analysis theoptimised simulation parameters presented in Table 7.1 were used with the followingexceptions:• The upstream data rate of 3.088 Mbps was changed for an upstream channel with acapacity of 6.176 Mbps. This change was considered necessary because the resultspresented in the previous sections and also in the previous two chapters were basedon a 3.088 Mbps upstream channel. The DVB/DAVIC protocol is also capable ofsupporting a 6.176 Mbps in the upstream direction. Therefore, in order to approachthe maximum system performance in terms of the maximum number of activestations supported, an upstream channel with a capacity of 6.176 Mbps was used.• The number of signalling frames transmitted in the 3 ms period was changed to 4,instead of 1. This is because stations can achieve a faster interaction (as examined inSection 5.4.6) with the headend, if more than one signalling frame is transmittedwithin the 3 ms period, which results more convenient for the support of TCIS.• The minimum number of CSs allocated per signalling frame was changed to one,instead of two. In Section 6.2.2.2, it was proved that the maximum system7-18

Chapter 7Performance optimisation for the support of TCIS pperformance is yielded with at least one CS per signalling frame when the splittingtree algorithm is used.• The maximum number of active stations analysed was up to 680 instead of 180 inorder to cover the point when saturation is experienced and very low-data ratestreams are delivered.A) Performance analysis for mixed trafficIn this section, a performance comparison is provided when the PRA and QoS (with afixed-rate connection) mechanisms are used for the transmission of mixed traffic. Thesimulations were carried out using only the splitting tree algorithm and the mixed trafficconsisted of 9.7 kbps of VoIP traffic and 32 kbps of Internet traffic, as described inSection 5.3.3.It can be seen in Figure 7.15a that the delay characteristics for both traffic types weresimilar for the PRA mechanism, which supports approximately 80 stations. On the otherhand, with QoS not only an extended number of VoIP connections can be supported (upto 320), but also the mean access delay is reduced and ranged between ≈ 1 or 2 msbefore the system becomes unstable.This reduction in access was because the number of MCI frames transmitted in the 3msperiod was increased to 4 and this helped the NIUs to achieve a faster interaction withthe INA, and also because VoIP traffic had precedence over IP traffic at the headend.a) Mixed Traffic: Mean Access DelayMean Access Delay (ms)50454035302520151050Internet & VoIPTrafficPRAInternet TrafficQoS-FxdrateVoIP Traffic40 80 120 160 200 240 280 320b) Mixed Traffic: System ThroughputSystem Throughput (%)5550454035302520151050Internet TrafficPRAPRAQoS-FxdrateVoIP TrafficQoS-Fxdrate40 80 120 160 200 240 280 320umber of Active Stationsumber of Active Stations7-19Figure 7.15 – Mean access delay and system throughput vs. No. of active stations.Splitting tree algorithm with mixed traffic.

Chapter 7Performance optimisation for the support of TCIS pFrom Figure 7.15b, the maximum sustainable system throughput was ≈ 56% of the ccfor the PRA mechanism. For the mechanism provided by QoS, the maximum systemwas also ≈ 56%. Internet traffic was choked off because in every MCI frame, theheadend schedules first the bandwidth for VoIP traffic and then in the remaining slots IPtraffic. This becomes more evident when the maximum system throughput is achieved(at 56% of the cc, produced by 80 stations).B) Performance analysis for VoIP trafficIn this section the performance is focused on the case in which a fixed-rate connectionis used for the delivery of VoIP streams. Here two different audio codecs are used forthe analysis, G.711 [56] and G.723.1 [57]. So far, most of the VoIP streams examinedhave been based on the codec G.723.1, which encapsulates 4 VoIP frames of 20 bytesevery 120 ms. In order to approach the maximum system performance that theDVB/DAVIC protocol can achieve, we have used four possible configurations for eachcodec. In addition, to increase the number of streams supported, header suppression(which is an advanced functionality in the latest version of this protocol [35]) is alsoconsidered in this analysis.Codec G.711 was considered here to stress the CATV network and also becauseEuroModems (class B) are more likely to use this codec for quality voice calls. G.711 isthe mandatory codec according to the ITU-T H.323 conferencing standard [59], whichuses Pulse Code Modulation (PCM) to produce a data rate of 64 kbps. This audio codeccreates and encapsulates a VoIP frame of 80 bytes every 10 ms. In order to reduce theprotocol overhead involved, a frame size of 30 ms (240 bytes) can also be used,according to the Packet Cable Audio/Video Codecs Specification [80].7-20

Chapter 7Performance optimisation for the support of TCIS pFrom Table 7.2 (second and fourth column) we can see that without header suppression(HS) only 55% (for a frame size of 10ms) and 78% (for a frame size of 30ms) is voicedata. According to [47], there is interest in the cable modem communities in removingthe requirement for LLC encapsulation of voice streams at the MAC layer, which wouldincrease the fraction of each voice data in each packet to 63% and 83%, respectively.Further, header suppression involving the RTP, UDP and IP headers would result ineven more significant bandwidth savings. In the simulations it was assumed that theRTP, UDP and IP headers can all be suppressed, which is reasonable if the INAmaintains additional state information on all active voice connections [47]. Thus, theefficiency calculated from Table 7.2 (third and fifth column) is considerably improvedto 91% and 97% for frame sizes of 10 and 30 ms, respectively.Similarly for G.723.1, four configurations were also analysed as indicated in the lastfour columns of Table 7.2. The frame sizes for this codec were of 30 and 120 ms.Table 7.2 – VoIP codecs: G.711 and G.723.1G.711 - 64 kbps (ISD) G.723.1 - 5.3 kbps (Internet)without HS with HS Without with HS without HS with HS without HS with HSFrame Size [ms] 10 10 HS 30 30 30 30 120 120Voice Frame [bytes] 80 80 240 240 20 20 80 80RTP [bytes] 12 0 12 0 12 0 12 0UDP [bytes] 8 0 8 0 8 0 8 0IP [bytes] 20 0 20 0 20 0 20 0LLC [bytes] 3 3 3 3 3 3 3 3SNAP [Bytes] 5 5 5 5 5 5 5 5Ethernet MAC [bytes] 18 0 18 0 18 0 18 0Voice Packet size [bytes] 146 88 306 248 86 28 146 88Net Data rate [kbps] 116.8 70.4 81.6 66.1 22.9 7.5 9.7 5.97-21

Chapter 7Performance optimisation for the support of TCIS pa) G.711 PerformanceResults presented in Figure 7.16a revealed that the maximum number of streamssupported with a frame size of 10 ms is about 27 stations, which achieves a maximumsystem throughput at approximately 50% of the channel capacity (Figure 7.16b).Increasing up to three the number of frames per audio packet not only results in a gainof about 10% on system throughput, but also in an increased number of VoIP streamssupported (up to 45 stations).The maximum system capacity is yielded when header suppression is considered,achieving up to 53 VoIP connections, regardless of the frame size. However, themaximum system throughput of about 60% of the cc is only produced with a frame sizeof 10 ms. The other frame size achieves a reduced capacity at approximately 56% of thecc. This is because with a frame size of 30 ms, the VoIP packet size to be transmitted isof 248 bytes that requires an additional (8-byte) AAL5 header plus 32 bytes of paddingstuff. This extra padding is used to get an entire AAL5-PDU multiple of 48 bytes, whichresults in a waste of bandwidth and degrades the system throughput as indicated inFigure 7.16b.Mean Access Delay (ms)109876543210a) G.711: Mean Access Delay b) G.711: System ThroughputFs-10msFs-30msFs-30ms+HSFs-10ms+HS15 20 25 30 35 40 45 50 55umber of Active StationsSystem Throughput (%)646158555249464340Fs-30msFs-10ms+HSFs-30ms+HSFs-10ms40 43 46 49 52 55 58 61 64Offered Load (%)Figure 7.16 – Mean access delay and system throughput.Splitting tree algorithm and VoIP traffic (G.711 at 64 kbps).7-22

Chapter 7Performance optimisation for the support of TCIS pb) G.723.1 PerformanceResults for the codec G.723.1 (presented in Figure 7.17a) show that by using a framesize of 30 ms, up to 160 stations can be sustained at 59% of the cc. This number can beextended to 320 VoIP connections with header suppression, but a reduction in terms ofsystem throughput is obtained, achieving about 39% of the channel capacity. This loss(of about 20% of the cc) is caused by the 12 bytes padding used to encapsulate the 28-byte audio packet to fit into one (48-byte) AAL5 PDU (as indicated above). The otherthree Voice packet configurations do not suffer this large reduction, because they fitproperly into a fixed number of ALL5-PDUs.With codec G.723.1, optimum system performance is obtained when header suppressionis considered and a frame size of 120 ms is used. The network capacity increases to thehighest figure for this research supporting up to approximately 640 VoIP connectionsand producing a mean access delay below 2 ms before congestion is experienced(Figure 7.17a). This equates to 60% of the available channel capacity (Figure 7.17b).The results presented in this subsection for both audio codecs can be validated by usingEquations 7.1 and 7.2, for the maximum system throughput sustainable and themaximum number of streams supported, respectively.Mean Access Delay (ms)109876543210a) G.723.1: Mean Access DelayFs-30msFs-120ms+HSFs-120msFs-30ms+HS0 100 200 300 400 500 600 700umber of Active StationsSystem Throughput (%)636057545148454239363330b) G.723.1: System ThroughputFs-120ms+HSFs-120msFs-30ms+HSFs-30ms30 35 40 45 50 55 60 65Offered Load (%)Figure 7.17 – Mean access delay and system throughput.Splitting tree algorithm and VoIP traffic (G.723.1 at 5.3 kbps).7-23

Chapter 7Performance optimisation for the support of TCIS p⎛ Pk ⎞size ⎛ RS ⎞S⎜⎟max= CC ⋅ ⋅ ⎜ ⎟(7.1)⎝ Pkslots⎠ ⎝ RS + CS ⎠Smaxmax=(7.2)⋅ λPk sizewhere RS and CS corresponds to the number of slots to be allocated to the reservationand contention access respectively. For instance, for the codec G.723.1 with a framesize of 30 ms (20 bytes of sampling), without header suppression the audio packet sizeto be transmitted becomes of 86 bytes (see Table 7.2, fifth column). Then, fromEquations 7.1 and 7.2,⎛ 86 ⋅ 8 ⎞ ⎛ 8 ⎞Smax= 6176000 ⋅ ⎜ ⎟ ⋅ ⎜ ⎟ = 3688 kbps = 59.7% of the cc⎝ 2 ⋅ 64 ⋅8⎠ ⎝ 8 + 1⎠maxS=maxSmax 3688000= =≈ 161⋅ λ (86 ⋅8)⋅ (1/ 0.03)rP sFrom simulations using this codec, a maximum system throughput of 59% of the cc anda maximum number of streams supported of 160 was obtained. Thus, theoretical resultswere seen to be similar. In general, for all four configurations presented for each codeca maximum deviation between simulation results and theoretical results was found to beless than 1.5%. This deviation can be attributed mainly to collisions and protocoloverheads.7.4 Comparison: DVB/DAVIC vs. DOCSISSo far only a performance characterisation and optimisation for the European cablecommunication system ‘DVB/DAVIC’, under various configurations and differenttraffic situations has been introduced. However, as stated in Section 2.5, theEuroDOCSIS protocol is also a serious alternative for the European market.This section continues with a comparison between both standardised cablecommunication protocols: DVB/DAVIC vs. DOCSIS. The comparison focuses onperformance issues and technical characteristics at the MAC and PHY layers.7-24

Chapter 7Performance optimisation for the support of TCIS pIn Barton [8], a comparison of the main characteristics of each protocol was providedfrom an American perspective. Here, an overview and status of each standard isprovided. Furthermore, a technical comparison of the major characteristics, strengths,advantages and disadvantages of each standard are also described.A less extensive comparison but with a European perspective is reported in [55]. In thispaper the authors presented an overview of each protocol followed by a briefcomparison of the major technical characteristics of each standard.An initial performance comparison carried out in this research can be found in [84].This comparison is based on the DVB/DAVIC, DOCSIS and the (withdrawn) IEEE802.14 standard. The comparisons presented in this section focus on services, MAClayer, PHY layer technical characteristics, and performance comparisons in terms ofvolumetric data (at the physical layer), efficiency, mean access delays and maximumsystem throughput.7.4.1 Technical comparisons7.4.1.1 MAC layer comparisonsAt the DVB/DAVIC MAC layer, four access modes are provided. The first mode(contention access) is based on a contention access mechanism, which allows users tosend information at any time with the risk of collisions. The second and third modes(fixed-rate and reservation access) are contentionless, in which the INA either providesa predefined amount of slots to a specific NIU, or a given bit rate requested by an NIUuntil the INA stops the connection on NIU's demand. These access modes aredynamically shared between time slots, which allows the NIUs to know whencontention, reservation or fixed-rate transmission is or is not allowed. This is to avoid acollision for the two contention-less based access modes. The fourth mode is calledranging access, in which slots are used to measure and adjust the time delay and thepower level for upstream transmission.Data transmission in the DOCSIS MAC layer is based on a request/grant scheme. Atregular intervals, the CMTS provides timing, ranging, registration, transmission andcontention resolution information for every CM on the network. In the upstream, each7-25

Chapter 7Performance optimisation for the support of TCIS pminislot can be used either for contention, contention/data, initial maintenance or stationmaintenance. A CM with data to send issues a request, using contention minislots. Inthe case of a collision, this protocol applies the exponential backoff algorithm and theinitial and truncated backoff values of this algorithm are indicated in each signallingframe called as MAP. This frame also describes the usage of the upstream bandwidth.7.4.1.2 PHY layer comparisonThe basic characteristics at the physical layer are summarised in Table 7.3. In this tablethe downstream and upstream spectrum allocations and the slight variations in thefrequency ranges for the DVB/DAVIC and DOCSIS standards are shown. Thedownstream channels support the legacy analogue broadcast television (frequency rangeare ≈ 80-450 MHz) and multiples of 1-6 MHz (for DOCSIS) or 1-8 MHz channels (forDVB/DAVIC and EuroDOCSIS) in the 450-860 MHz region for transmission of digitaldata. The upstream channels are also divided into 1-6 MHz but poor SNR limits the datacapacity to 1-10 Mbps per channel for DOCSIS and 1-6 Mbps for DVB/DAVIC. This isa low data rate compared to the downstream channel where 28-40 Mbps and 42-52Mbps are available for DOCSIS and DVB/DAVIC, respectively.In terms of modulation schemes, most manufacturers have implemented 64 and 256-QAM for the downstream and QPSK for the upstream channels. DOCSIS also supportsa 16-QAM modulation in the upstream direction.For upstream transmission at the physical layer, DOCSIS MAC PDUs (which arecomposed of Ethernet packets) are segmented into PHY codewords, which are ofTable 7.3 – Frequency allocation and modulation characteristics.UpstreamDownstreamFeature DVB/DAVIC DOCSISModulation QPSK, 16-QAM under development QPSK, 16-QAMFrequency 5 – 65 MHz DOCSIS: 5–42MHz, EuroDOCSIS: 5-65MHzSpacing 200 kHz, 1 MHz, 2 MHz, 4 MHz 200, 400, 800 kHz, 1.6, 3.2MHzData rates Mandatory: 3.088 Mbps,320 kbps, 640 kbps,Optional: 256 kbps, 1.54 kbps, 6.176 Mbps 1.28 Mbps, 2.56 Mbps, 5.12 Mbps, 10.24 kbpsModulation IB: QPSK, OOB: 16, 32, 64, 256-QAM 64-QAM or 256-QAMFrequency OOB: 70-130 MHz / 300-862 MHz,80 – 860MHzIB: 300-862 MHzSpacingOOB: 1 MHz, 2 MHz,DOCSIS: 6MHz, EuroDOCSIS: 8 MHzIB: 7/8 MHzData rates 52 Mbps (256-QAM), 42 Mbps (64-QAM) 42 Mbps (256-QAM), 30 Mbps (64-QAM)7-26

Chapter 7Performance optimisation for the support of TCIS pvariable size (ranging from 16 to 253 bytes). These codewords contain FEC parity, alsoof variable size (from 0 to 10 bytes). Both a preamble and guard time fields of variablelength are then added at the beginning and at the end of all the codewords forsynchronization purposes between the CM and the CMTS. Such codewords aretransmitted as a continuous series of minislots. The size of the minislot is set by theCMTS during initialisation but can be varied. The length of a minislot is a multiple of6.25µs (i.e. 8, 16, 32 bytes etc.).In the upstream, DVB/DAVIC uses a 64-byte slot format for the transmission of data atthe physical layer. The slot format consists of a Unique Word which provides a burstmode acquisition method, a payload area that contains a single ATM cell and a Reed-Solomon parity field [34], which provides 3-bytes of FEC over the payload area and aGuard Band field for synchronization. The basic characteristics of these two protocolsare outlined in Table 7.4.Table 7.4 – Technical characteristics.Feature DVB/DAVIC DOCSISServicesInternet access, high speed interactive Settop-box,Internet access, low speed interactive Set-top-VoIP, SNMPbox, VoIP,SNMPUpstream Packet Format 64-byte Slot based on a ATM cell transport Variable Length (based on 64-1500 byteswith IP adaptation layer translationEthernet packets), Native IP with QoSQoS Granularity 53 bytes + 11 bytes PHY overhead 8-16 bytesQoS ServicesCollision ResolutionBE plus ATM derived class of services:CBR, ABR.Exponential backoff and Splitting treealgorismBE, CIR and prioritisation in DOCSIS 1.0 andRTP, nrt-RTP, UGS added in DOCSIS 1.1Exponential backoff algorithmATM Support Mandatory OptionalAccess Modes Ranging, contention, reservation and fixed Contention and reservationSecurity RSA/DES Encryption and clone detection RSA/DES Encryption and clone detectionCommercial Deployment DVB/DAVIC EM: Q3-99, Q3-00. DOCSIS 1.0: Q2-99; DOCSIS 1.1 Q4-997.4.2 Performance comparisonsThe performance comparison to follow addresses some of the fundamental propertiesand scalability of the upstream channel for both protocols. The analysis focuses on theperformance comparison for the upstream channel, which is the limiting factor onCATV networks (as stated in Section 1.2.2), and is critical in the delivery of services toindividual subscribers on demand. Key issues of the analyses address the followingperformance comparisons: protocol efficiency and volumetric data at the physical layer,7-27

Chapter 7Performance optimisation for the support of TCIS pmaximum throughput achieved per cable modem, effects of varying the packet size inisochronous streams and maximum system capacity in terms of active cable modems.For the DOCSIS protocol, the CSF (v.13) based on the OPNET simulation package wasused [77]. The simulation parameters considered for the performance analysis aresummarised in Table 7.5. In both protocols a scheduler at the headend with a simpleFIFO service was used. For the DOCSIS protocol, priority service was not considered,since the DVB/DAVIC standard does not support traffic prioritisation. Therefore, allstreams generated are assumed to have the same access priority.In addition, the default upstream data rate of 3.088 Mbps for the DVB/DAVIC protocolwas selected. The corresponding data rate for the DOCSIS protocol was set at 2.056Mbps. The concatenation and piggyback features of the DOCSIS protocol were notconsidered for this analysis, since such characteristics are not supported by theDVB/DAVIC protocol.Table 7.5 – Simulation parameters.Parameter DOCSIS DVB/DAVICUpstream channel capacity (QPSK) 2.56 Mbps 3.088 MbpsNumber of cable modems Up to 350 Up to 350Minislot/slot size in bytes 16 (minislot) 64 (slot), 21 (minislot)Max. No. of minislots/slots in MAP/MCI Max. 4096 Fixed 36Min. No. of contention slots in MAP/MCI 16 [2-3]Contention resolution algorithm Exponential backoff algorithm Splitting tree algorithmMax. number of IEs in MAP 240 ---------FEC-T bytes 3-SD, 5-LD* 3Codeword length 75-SD, 245-LD* 53 (ATM cells)Last codeword shortened True ---------Guard time size 2 bytes (variable) 1 byte (Fixed)Distance from nearest/farthest modem to the headend 10-16Km, rnd. distributed 10-16Km. rnd. distributedSimulation time 60 seconds 60 seconds*SD = Short data codeword, LD = Large data codewordA) Protocol efficiency and volumetric data comparisonsThis analysis addresses a comparison of the volume of data (transferred upstream by auser data request transmission) and protocol efficiency for the DVB/DAVIC and theDOCSIS standards.7-28

Chapter 7Performance optimisation for the support of TCIS pFor the upstream volumetric data (Figure 7.18a), the DVB stepped curve is caused bythe encapsulation of PDU messages into ATM cells (of 53 bytes), which are thentransmitted throughout the upstream channel as 64-byte frame slots. Therefore, a usermessage of 1450 bytes (without protocol overhead) causes the DVB protocol to requestthe following data: (1450-payload + 20-TCP + 20-IP + 3-LLC + 18-MAC + 8AAL5) /48 AAL5 PDUS ≈ 32-slots = 2048 bytes. This results in 71 % upstream efficiency asindicated in Figure 7.18b. Here, every upstream slot transmitted includes 6 bytes of FECinformation to correct up to 3 bytes (T=3) over each ATM cell. Figure 7.19 presents theupstream PDU structures of this protocol.Volumetric Data (bytes)2250200017501500125010007505002500a) Upstream Volumetric Data204815901520DOCSIS, no FECDVBDOCSIS, with FEC0 250 500 750 1000 1250 1500User Message Length (byes)Efficiency (%)1009080706050403020100b) Upstream EfficiencyDOCSIS->95%, no FECDVB->71%DOCSIS->91%, with FEC0 250 500 750 1000 1250 1500User Message Length (bytes)Figure 7.18 – Upstream volumetric data and efficiency.1450-byte user dataApplication byte streamTCP-Hdr.20-byte1450-byte user dataTCP Segment (TCP-PDU)IP-Hdr.20-byte1470-byte TCP PDUIP Datagram (IP-PDU)LLC-Hdr.3-byte1490-byte IP PDULLC Segment (LLC-PDU)MAC-Hdr.14-byte1493-byte IP PDUMAC-FCS.4-byteMAC Frame (MAC-PDU)DVB-Hdr.9-byteATM payload48-byte64-byte Upstrem Slot (1)DVB-Tr.7-byte1511-byte MAC PDUDVB-Hdr.9-byteCPCS-Tr.8-bytesATM payload31-bytePAD17-byte64-byte Upstrem Slot (32)ALL5-CPCS (CPCS-PDU)DVB-Tr.7-byteUpstream SlotDVB/DAVICMAC&PHYFigure 7.19 – DVB/DAVIC upstream PDU structures.7-29

Chapter 7Performance optimisation for the support of TCIS pSimilarly, the DOCSIS stepped curve is caused by the encapsulation of PDU messagesinto codewords (of ≈ 220 bytes), which are then transmitted as a continuous series ofminislots. Figure 7.20 shows the PDU structures for this protocol. For instance, for thesame user data of 1450 bytes, the DOCSIS protocol requests (1450-payload + 20-TCP +20-IP + 3-LLC + 18-MAC + 6-DOCSIS MAC) / 245 byte-codeword = 6 largecodewords + 1 short codeword ≈ 1590 bytes (or ≈100 minislots), for an upstreamefficiency of 91%. Here, every codeword includes 10 bytes of FEC information tocorrect up to 5 bytes over the MAC PDU. Without FEC the volumetric data at the PHYlayer becomes of 1520 bytes (≈95 minislots) for an increased efficiency up to 95%.MAC-Hdr.14-byte1493-byte IP PDUMAC-FCS4-byteMAC Frame (MAC-PDU)PreambleDOCSIS MAC6-bytes220-byte Packet DataFEC220-byte Packet Data1511-byte IMAC PDUFEC197-byte Packet DataFECDOCSIS MAC FrameGuradTimeCodewordsDOCSISMAC&PHY10-bytes 230-byte FEC Codeword (1)230-byte Codeword (2)207-byte FEC Codeword (7) 1-byteFigure 7.20 – DOCSIS upstream PDU structures.B) Maxim throughput achieved per cable modemIn order to obtain the maximum throughput per cable modem that each protocol canachieve, a single node (cable modem) network was considered with typical Ethernetpacket sizes ranging from 64 to 1518 bytes. A traffic generator within the node was setto produce the maximum number of packets that it could receive from the CPEinterface, (as introduced in Section 5.4.7.1), which was a 10 Mbps constant stream withvariable number of packet per second depending on the packet size.All references to packet size in this analysis (and also in the following analyses) willrefer to the size of the packet as it enters the system from the CPE interface. This wouldbe the packet size that the cable modem submits for delivery over the cable network,which does not include CATV MAC and PHY overheads. However, the CPE MAC andpossible layer 2 and layer 3 overheads are included.7-30

aChapter 7Performance optimisation for the support of TCIS pThe maximum throughput and upstream channel utilisation versus the packet sizegenerated is shown in Figure 7.21. In terms of throughput (Figure 7.21a), the DOCSISprotocol is superior to the DVB/DAVIC protocol. For the maximum Ethernet packetsize (1518 bytes), the maximum sustainable throughput per DOCSIS cable modem was≈ 1.6 Mbps, which corresponds to about 62% of the upstream channel capacity, incomparison to ≈ 1 Mbps (32% of the cc) achieved by each DVB/DAVIC cable modem.In terms of utilisation (Figure 7.21b), the difference was 19% of the link capacity infavour of the DOCSIS protocol.However, for small packet sizes under 256 bytes, results reveal that bothcommunication protocols are highly inefficient. The maximum throughput does notexceed 25% of the link capacity even though the offered load was a constant 10 Mbpsstream. The worst case scenario is released for the 64-byte packet size, where thethroughput was as low as 2% of the cc (≈ 56 kbps) for DVB/DAVIC cable modems, incomparison to 7% of the nominal channel capacity (≈ 181 kbps) provided by theDOCSIS cable modems.A closer analysis of the difference between throughput and utilisation, simulation resultsshowed that an increased protocol overhead is involved in the DVB/DAVIC protocolthat reduce considerably the protocol efficiency, as examined in the previous scenario.Throuhgput (Mbps)2.01.81.61.41.21.00.80.60.40.20.0a) System Throughput b) UtilisationDOCSIS: 2.5Mbps UpDVB: 3Mbps Up64 128 256 512 768 1024 1280 1518Packet Size (bytes)Utilisation (Mbps)2.01.81.61.41.21.00.80.60.40.20.0DOCSIS: 2.5Mbps UpDVB: 3Mbps Up64 128 256 512 768 1024 1280 1518Packet Size (bytes)Figure 7.21– Maximum channel capacity for DVB/DAVIC and DOCSIS for one node.7-31

Chapter 7Performance optimisation for the support of TCIS pFor instance, for packet sizes of 1518 bytes, this difference resulted in approximately2% and 13% of the cc for the DOCSIS and DVB/DAVIC protocols, respectively.C) Maximum network capacityIn this analysis, we now address the maximum scalability of the network in terms ofmean access delays and system throughput/utilisation for an increased networkpopulation. The number of stations was increased to 340 to move beyond saturationpoints. The traffic load generated by each station was a single-Ethernet packet of 1518bytes with an exponentially distributed inter-arrival rate of 1 packet per second. This isequivalent to 12 kbps streams.Results presented in Figure 7.22a tell us that both protocols can support the samenumber of stations, (at least in this traffic situation) with approximately 160 streamssupported, after which delays become too large for the chosen application. It is apparentthat for medium network traffic loads, DVB/DAVIC cable modems provide slightlylower delays than the DOCSIS protocol, due to the faster interaction with the headend.DVB/DAVIC always sends an MCI frame every 3 ms and DOCSIS uses a variablescheme, where the number of minislots described per MAP can be from a few minislotsMean Access Delay (ms)240220200180160140120100806040200a) Mean Access Delay b) Channel CapacityDOCSISDVB20 60 100 140 180 220 260umber of Active StationsChannel Capacity (Mbps)3.02.72.42.11.81.51.20.90.6UtilisationThroughputDVBDOCSISDVB20 60 100 140 180 220 260 300 340 380umbe r of Active StationsFigure 7.22 –Mean access delay and maximum channel capacity,for DVB/DAVIC and DOCSIS, 12.4 kbps streams with packet size = 1518 bytes.7-32

Chapter 7Performance optimisation for the support of TCIS pup to 4096. This results in MAP being transmitted from ≈ 1ms when only contentionminislots are described up to (4096minislot·128bits/2560000≈) 200 ms on high trafficloads.Figure 7.22b shows an increased system performance for DOCSIS over theDVB/DAVIC protocol in terms of system throughput and utilisation, even though areduced upstream channel capacity was used for the DOCSIS protocol. For thisanalysis, the maximum system throughput resulted in ≈ 65% of the channel capacity forthe DVB/DAVIC protocol, in comparison to 90% of the cc achieved by the DOCSISprotocol.In terms of utilisation, the DVB/DAVIC protocol uses 89% of the channel capacity incomparison to 93% utilised by the DOCSIS protocol. Both protocols cannot achieve100% of the upstream channel capacity due to slots assigned to the contention accessregion that remained unused. Reducing the minimum number of contention slots perMAP or signalling frame may result in an increase of upstream channel utilisation, butthere may be also some drawbacks, such as an increased number of collisions in periodsof congestion and therefore an increased mean access delay.D) Effects of packet size in isochronous streamsHere, an analysis of the performance impact in terms of mean access delay is presented.The system throughput provided by each protocol when the packet size is ranged from64 to 1518 bytes for different isochronous streams is shown. For this analysis, streamsfrom 16 kbps to 128 kbps, suitable for low-rate timing critical interactive services suchas compressed/ uncompressed voice, audio and low quality video were considered.7-33

D O C S I SChapter 7Performance optimisation for the support of TCIS pFigure 7.23 shows the performance comparison when isochronous streams at 32 kbpswere used for small, medium and large packet sizes.From the results for a 64-byte packet size, in Figure 7.23a it can be seen that themaximum number of streams that the DVB/DAVIC protocol can support isapproximately 24, with access delays under 20 ms. The DOCSIS protocol supportsabout 8 streams more than the number achieved by DVB/DAVIC. Results for the othertwo packet sizes (256 and 1024 bytes) show that both protocols support the samenumber of streams for delay sensitive isochronous streams (under 20m), whichcorresponds to 54 and 57 streams for 256 and 1024-byte packet sizes, respectively. Forsome isochronous streams that support larger access delays over 20 ms, e.g. video,audio, the number of streams sustainable resulted in a large figure for the DOCSISprotocol.In terms of system throughput, (see Figure 7.23b) the minimum Ethernet packet sizeyields a maximum system throughput of 1.4Mbps (55% of the cc) for the DOCSISprotocol and 1 Mbps (32% of the cc) for the DVB/DAVIC standard, which indicatesthat the DOCSIS protocol is capable of using approximately 23% of the upstreamchannel capacity for the delivery of more user data. For packet sizes of 256 and 1024bytes, this difference resulted in 25% and 28% in favour of the DOCSIS protocol.Mean Access Delay (m s)1 9 51 8 01 6 51 5 01 3 51 2 01 0 59 07 56 04 53 01 50a) Mean Access DelayD V B64byte-PkDOCSIS1 8 2 4 3 0 3 6 4 2 4 8 5 4 6 0 6 6 7 2D V B256bytePkDOCSISD V B um be r o f A c tiv e Statio nsDOCSIS1024bytePkSystem Throughput (Mbps))2 .52 .32 .11 .91 .71 .51 .31 .10 .90 .70 .5b) System ThroughputDOCSIS256 byte PkD V B64bytePkDOCSIS1024bytePkD V BDOCSISD V B1 8 2 7 3 6 4 5 5 4 6 3 7 2 8 1 9 0 9 9 umbe r o f A ctiv e Statio nsFigure 7.23 - Mean access delay and system throughput, for DVB/DAVIC and DOCSISwith 32 kbps isochronous streams and different packet sizes.7-34

Chapter 7Performance optimisation for the support of TCIS pResults for the complete range of packet sizes and isochronous streams are summarisedin Table 7.6. These results are displayed in terms of maximum number of streamssupported for both protocols. For packet sizes from 64 to 512 bytes, access delays under20 ms were considered. Meanwhile for larger packet sizes, (1024 and 1518 bytes)access delays under 100 ms were used, since it is more likely that larger delays areaccepted for the support of isochronous streams when large packet sizes are utilised.In general, for most of the packet sizes and isochronous streams evaluated, DOCSISoutperformed the DVB/DAVIC protocol. There were just few cases where theDVB/DAVIC supported an increased number of streams, as it was the case of low rateisochronous streams of 16 kbps and 32 kbps for the packet size of 128-bytes.For higher data rate isochronous streams and small packet sizes, there were threeinstances where the DVB/DAVIC protocol failed to provide low transmission delays(under 20 ms). This is because according to the CRGC of the DVB/DAVIC protocol(depicted in Figure 3.3) the minimum packet delay is approximately of the order of 9ms, and the inter-arrival time for the three cases indicated in Table 7.6, were either 8 msor 4 ms. This resulted in infinite queue built-up at the transmitter side causing thestreams to fail. In order to fix this problem, enhanced reservation request mechanismsare required, as introduced in Section 7.2 and [87]. Another possible solution is todecrease the transmission cycle of the DVB/DAVIC signalling frame, as analysed inSection 5.4.6.Table 7.6 – Number of streams supported for DVB/DAVIC and DOCSIS.PacketIsochronous StreamsSize Access 16 kbps 32 kbps 64 kbps 128 kbps(bytes) Delay(ms) DVB DOCSIS DVB DOCSIS DVB DOCSIS DVB DOCSIS64 ≤20 62 62 24 32 No supported 20 No supported 6128 ≤20 96 87 48 43 19 23 No supported 12256 ≤20 108 110 54 54 27 27 12 14512 ≤20 112 125 59 62 31 31 15 151024 ≤100 122 143 61 70 31 35 15 181518 ≤100 125 144 63 71 31 36 15 187-35

Chapter 7Performance optimisation for the support of TCIS pAs a final remark, results in terms of maximum number of streams supported for a6.176 and 5.12 Mbps upstream channel can be found in [85] and [121] for theDVB/DAVIC and the DOCSIS protocol, respectively.7.5 ConclusionsIn this chapter several novel improvements were implemented that would give theDVB/DAVIC MAC protocol a superior performance and increased efficiency for thesupport of the next generation of communications services.The first improvement was through the use of six enhanced-reservation-requestmechanisms. As these mechanisms become more complex, they efficiently reduce andin some cases avoid the collision risk of reservation request transmissions, increasingmarkedly the system performance by reducing the access delay and achieving a highersystem throughput. Results presented have shown that the basic reservation accessmechanism of the DVB/DAVIC standard can be improved upon significantly byadopting these enhanced mechanisms. The unsolicited grant slot mechanism (when theCBR parameters are satisfied with regard to the actual data rate, packet size and packetinterval) and the continuous piggyback mechanism are the most appropriate for thesupport of timing critical interactive services. However, the use of minislots with orwithout piggybacking requests, and the enhanced pure reservation access mechanismsare best suited for Internet traffic or bursty traffic. The performance increase providedby these mechanisms can only be obtained when the traffic characteristics are carefullyanalysed. The choice of a poor mechanism may reduce the service quality considerably.The second approach for increased performance was through the use of Quality ofService. Here, the use of traffic prioritisation was found to have a significant effect onsystem performance. By giving higher transmission priority to isochronous streams thanto data packets, low packet access delays can be sustained (in medium and in somecases in high traffic loads) achieving an increased number of delay-sensitive streams.The major drawback of this prioritisation technique is that it does not provide anyguarantee of bandwidth availability or latency. However, in the reservation technique,through the use of a fixed-rate connection, a reduced packet access delay for the7-36

Chapter 7Performance optimisation for the support of TCIS pdelivery of isochronous streams, in the range of 1 and 2 ms could be guaranteed by theDVB/DAVIC protocol before large periods of congestion are experienced.Combined with header suppression, bandwidth efficiency is increased to a large extent,achieving a much higher figure regarding the maximum number of streams sustainable.For instance, for VoIP streams (with codec G.723.1) it was found that by using headersuppression and a large frame size of 120 ms, the maximum number of streamssupported could be increased up to approximately 640 connections, in comparison to160 when these features are not considered.In this research a minimum packet access delay of about 1ms (for VoIP streams) waspossible. However, with advanced synchronisation techniques this delay can be furtherreduced (under 1ms), for the delivery not only of VoIP packets but TCIS in general.Finally, a comparison between the DVB/DAVIC and the DOCSIS protocol was alsoaddressed in this chapter in order to evaluate the major properties provided by eachstandard at the MAC and PHY layers, which may help Internet Service Providers orCable Network Operators to make the right choice when performance issues for thesupport of timing critical interactive applications, and high-speed Internet traffic areconsidered.The DVB and the DOCSIS organisations are each producing their own cable modemspecification. At the PHY layer these specifications are similar. However, at the MAClayer the solutions provided by these groups have little in common. In general, in mostof the analyses, the DVB/DAVIC protocol achieved a reduced performance comparedwith the DOCSIS protocol. This was mainly because the DVB/DAVIC protocolencapsulates every datagram into ATM-AAL5 cells and suffers an overhead penalty forSegmentation and Reassembly (SAR) in an attempt to provide a faster transmission forQoS over the ATM protocol, while the DOCSIS uses a scheme that favours the deliveryof variable length Internet protocol packets rather than ATM transfer, in an attempt tokeep cost and complexity of cable modems down.7-37

Chapter 8FINAL CONCLUSIONS8.1 IntroductionThe work presented in this dissertation has addressed the issues of digitalcommunications over the European cable communication protocol (DVB/DAVIC, ETS200 800). The focus of this work has been the performance evaluation and optimisationof the upstream channel, which is more complex to analyse when compared with thedownstream channel. This is because transmission on the downstream channel ishandled exclusively by the headend (the Interactive Network Adaptor), simplifyingoperation. In contrast, the upstream channel is a shared access medium, which usesrandom (contention), reservation and fixed access techniques. The interval distributionof the start time of these access modes is dynamic and controlled by the headend. Theboundaries of contention and reservation access are broadcast periodically in thedownstream channel. The boundary of fixed rate access is assigned to a station at thebeginning of connection. When collisions occur, a contention resolution algorithm isused to resolve them. This protocol uses two different CRAs, which are the exponentialbackoff algorithm and the splitting tree algorithm.The series of analyses presented in this dissertation have concentrated on theeffectiveness of the access modes and the CRAs defined in the DVB/DAVIC protocolspecification with the key performance issues for access and data transmission in theupstream channel. In addition, several improvements have also been introduced toextend this protocol in support of high-speed data transmissions and in particular timingcritical interactive services.8-1

Chapter 8Final conclusions p8.2 General discussionsChapter 1 presented an overview of the main characteristics of CATV networks. Hereit was highlighted that the bandwidth available in the upstream direction, makes CATVnetworks limited for the support of an increased number of users, particularly whentransmitting interactive services with high capacity. The chapter went on to give anoverview of the major measurements used in the field of performance analysis ofcommunication protocols, and provided a comprehensive review of the current researchin relation to stability of contention resolution algorithms and modelling of CATVprotocols relevant to this work. An important point here is that all contributions inresearch of performance analysis of CATV protocols have been made in the last sixyears.This chapter also discussed the contributions and novelty of this research. This projectwas started about two months after the first version of the DVB/DAVIC protocol cameout. With the exception of the work carried out in [100], most of the research found inthe literature focused on performance evaluation, improvements and optimisations forthe IEEE 802.14 and DOCSIS MAC protocols.In Chapter 2 the relevant theory was presented for this work in relation to CATVnetworks, which included a description of traditional and modern cablinginfrastructures, their differences and similarities. Here, the changes that have to be madeto upgrade CATV networks were pointed out for the support of bi-directional digitalcommunications.In addition, background material was also covered in this chapter that included anoverview of cable modem technology, possible cable modem configurations (internal,external or STB), cable modem providers and identified the requirements for the nextgeneration of communications services.This chapter also outlined the main characteristics of the DOCSIS, IEEE 802.14 andDVB/DAVIC and presented an overview of alternative technologies for high-speeddigital access, including xDSL, and FTTx infrastructures.8-2

Chapter 8Final conclusions pIn Chapter 3 a comprehensive description of the DVB/DAVIC protocol and the DVBproject was addressed. Here the main characteristics as well as the architecture, MACoperation, and a detailed description of the exponential backoff algorithm and thesplitting tree algorithm adopted by the DVB/DAVIC protocol specification weredescribed.Chapter 4 presented the simulation model that was implemented for the performanceevaluation of the DVB/DAVIC protocol. In summary, this model was based on theCommon Simulation Framework, initially developed by MIL 3 and CableLabs. Such aframework was implemented in the OPNET Simulation Package and is used for themodelling of the DOSCIS protocol. We have used the basic network topology of thisframework and programmed the functionality of the MAC and PHY layers of theDVB/DAVIC protocol (including the INA and NIUs), replacing the functionality of theMAC and PHY layers of the DOCSIS protocol, respectively.In addition, new features for further optimisations were also incorporated in thesimulation model. The major issues were the incorporation of three contention slotallocators, six enhanced reservation requests mechanisms, and a prioritised scheduleralgorithm at the station’s premises that were used in subsequent analysis.In this chapter, an analytical model was also formulated based on single node scenarioto validate the result that could be obtained trough the simulation model. A multi-nodeanalytical model was beyond the scope of this research, due to the complexity of theDVB/DAVIC protocol and its hybrid access method, which uses both random accesswith exponential backoff or splitting tree algorithm and elements of TDMA renders itstheoretical analysis very complex. We have found that studies of contention resolutionalgorithms are either based on simulation for accurate results or examine the stability ofthe algorithms with several simplifying assumptions, such as an infinite number ofstations.The simplified analytical model implemented for the DVB/DAVIC protocol was basedon an M/G/1 queuing system and addressed one of the basic performance properties ofcomputer communications protocols, which relates to the mean packet access delaysand maximum sustainable throughput of a station.8-3

Chapter 8Final conclusions pWe have seen that results obtainable using the simulation model were in goodagreement with theoretical results, with a maximum deviation of 3% for mean accessdelays and 2% for throughput. In this analysis it was found that the scheduler-lookahead is the major delay element in the transmission cycle, which dramatically reducesthe throughput.In chapter 5, before a performance optimisation or the incorporation of newmechanisms was approached, we first presented a rigorous performance evaluation ofthe fundamental properties of the DVB/DAVIC protocol, based on a 3 Mbps upstreamchannel.The first issue addressed was the protocol efficiency and performance characteristics interms of global offered load. Simulation results revealed that for the exponential backoffalgorithm, the protocol can sustain linear throughput increase and bounded delays foroffered loads up to 61% of the cc. The maximum system throughput was even higherand reached ≈ 65%. However, in order to achieve this throughput, the offered loadneeds to be higher that 70% of the cc at which point frame delay becomes unbounded.This throughput is higher when compared to other pure random access mechanisms,such as Aloha (37%), and CSMA/CD (60%), [117]. The main advantage is that thethroughput maintains maximum values even under heavy traffic loads and does notdegrade as rapidly as these other random access protocols.A similar performance characteristic with linear throughput and delay increase wasconfirmed in the following scenario studied (capacity in terms of active stations), wherean increased node population was considered. Here, the maximum sustainablethroughput was also 61% of the cc, with bounded access delays under 40 ms. Resultsfor the splitting tree algorithm revealed that this algorithm could achieve a highersystem performance than the exponential backoff algorithm, achieving a maximumsystem throughput of up to ≈ 65% of the cc.In addition, for these two analyses (offered load scalability and capacity in terms ofactive stations) the maximum channel utilisation achieved by the protocol was ≈ 90% ofthe channel capacity. The difference between maximum system throughput andutilisation is attributed to the extensive protocol overhead involved, generated by the8-4

Chapter 8Final conclusions pATM protocol and the DVB PHY layer, reservation request transmissions, collisionsand retransmissions. The remaining bandwidth of ≈ 10% was accounted for unusedcontention slots, due to the random nature of CRAs.The third analysis studied the delay and system throughput characteristics whentransmitting a determined number of cells in the ‘contention-based access region’, usingthe exponential backoff algorithm and a mixed traffic configuration of 32 kbps IP trafficand 12 kbps VoIP streams per station. In this analysis, it was found that optimumsystem performance is achieved when stations are allowed to transmit messages of 1-ATM cell in the contention access region. Here low delays under 25 ms were seen withan offered load up to 35% of the cc. Higher offered loads resulted in unbounded delays.In general, it was seen that when larger messages (e.g. 2-6 ATM cells) are transmittedusing contention access, the risk of collision is increased slightly, resulting in areduction in system throughput. This loss in throughput can be of 2% when messages up6-ATM cells are transmitted using contention access or up to 5% when all massages aretransmitted using only reservation access.Results for effects of maximum request size suggested that this parameter should be setas large as possible if the upstream channel is only used for Internet traffic. Forinstance, a value of 22 or 32 ATM cells was found to provide optimum systemperformance for this traffic configuration. However, if the upstream channel supportsthe transmission of both traffic types (IP and VoIP), a higher interaction for VoIPstreams can be obtained if the maximum request size is set as short as possible (e.g. 6ATM cells). Furthermore, when the performance of both CRAs was compared, it wasseen that the splitting tree algorithm could support up 10 stations more than the numbersupported with the exponential backoff algorithm. This is equivalent to a performanceincrease of up to 17% of the channel capacity.Results for the analysis of buffer capacity revealed that by using small buffer sizes, forexample of 50 or 100 ATM cells, only a small fraction, below ≈ 1% of the channelcapacity, is held in the station’s buffers on high traffic loads (above 53% of the cc),compared to over 10% of the cc when a large buffer capacity is used (e.g. 1000 or 3000ATM cells). Here, it was found that a small buffer capacity resulted in a better8-5

Chapter 8Final conclusions pperformance for the support of VoIP traffic, since lower access delays are obtained witha short buffer capacity than with a large buffer size. However, a drawback is that thesmaller the buffer capacity the higher the number of discarded packets, which mayresult in a degradation of service quality. For this analysis the number of discardedpacked resulted in ≈ 3 packets per second when a buffer capacity of 50 ATM cells wasconsidered. This number was of ≈ 1 and 0 for buffer capacities of 1000 and 3000 ATMcells, respectively.In general, from this analysis of buffer capacity, it was found that there is a naturaltrade-off between giving sessions free access to the network and keeping delay at a levellow enough so that interactive applications (e.g. VoIP, audio and video) are supportedand retransmissions or other inefficiencies do not degrade the network performance.From the analysis of the effects of increasing the number of signalling frames it wasfound that by transmitting two signalling frames in the 3 ms period, a slight decrease inaccess delay can be obtained. This reduction resulted in approximately 4 and 5 ms forthe exponential backoff algorithm and the splitting tree algorithm, respectively.However, in terms of system throughput, for the exponential backoff algorithm therewas a decrease in throughput of ≈ 6% of the cc. This reduction was because the numberof contention slots allocated in each signalling frame remained the same, reducing thebandwidth for data transmissions in every 3 ms cycle, in an attempt to resolve fastercollisions.The last analysis of this chapter addressed the effects of varying the packet size inisochronous streams. Here, it was shown that the major factors affecting the systemperformance were seen to be the length of the packet being transmitted for delivery. Inthe analysis of a single node scenario, it was demonstrated that regardless of the offeredload, a station cannot achieve throughput higher than 32% of the maximum channelcapacity in a 3.088 Mbps upstream channel. Even worse, this figure can be as low as1.8% of the cc when delivering minimum Ethernet packets, which is attributed to thescheduler-look ahead delay.8-6

Chapter 8Final conclusions pIn the multiple node scenario analysis, the DVB/DAVIC protocol proved again veryinefficient due to the excessive number of reservation requests, collisions,retransmissions and DVB MAC and PHY protocol overheads.Table 8.1 presents the maximum system performance and bandwidth characterisationsfor the six different packet sizes considered.In this scenario the maximum system throughput achieved was ≈ 34.5% of the cc whenpacket sizes of 64 bytes were considered. Increasing the packet size can significantlyimprove the performance with maximum system throughput up to ≈ 65% of the cc.Results for channel utilisation were higher than for throughput results with a deviationof 54% to 24% for minimum and maximum size Ethernet packets, respectively. Themajor source of performance inefficiency was the extensive protocol overhead involvedat the MAC (ATM encapsulation) and PHY (FEC) layers, which can be as high as 35%and 23% of the upstream channel capacity for 64-byte and 1518-byte packets,respectively. A second significant factor degrading system performance is thebandwidth consumed for the transmission of successful requests, which accounted for ≈12% when minimum size Ethernet packets are transmitted. Furthermore, collisions alsocontribute to system inefficiency with up to 8% of the cc.Chapter 6 introduced three novel contention slots allocators (Simple-CSA, Variable-CSA and Fixed-CSA). These techniques were studied because the DVB/DAVICprotocol specification did not define any mechanism for the allocation of contentionTable 8.1 – Summary of maximum system performance andbandwidth characterisation.BandwidthPacket Size (bytes)Characterisation (%) 64 128 256 512 1024 1518Maximum Throughput 34.5 51.3 56.2 62.9 63.4 64.6Maximum Utilisation 89.1 90.1 90.5 89.8 88.9 88.5Deviation 54.6 38.8 34.3 26.9 25.5 23.9Supported Streams 25.0 48.0 54.0 60.0 60.0 63.0Request Collided 8.3 4.1 1.3 0.4 0.1 0.1Request Successful 11.8 8.5 4.6 2.6 1.3 0.9MAC&PHY overhead 34.5 26.2 28.4 23.9 24.1 23.0Bandwidth unused 10.9 9.9 9.5 10.2 11.1 11.58-7

Chapter 8Final conclusions pslots into the signalling frames. Here it is most likely that unscheduled slots are assignedto contention access).In an analysis for the IEEE 802 protocol (reported in [94] and [96]) it was pointed outthat the performance of multi-access reservation protocols heavily depends on theoverall structure and the capacity assigned to the reservation and contention accessregions. Therefore, in this chapter we have demonstrated that the performance of theDVB/DAVIC can be improved if these techniques are used.The first strategy studied was the Simple-CSA. Here the minimum number of contentionslots per signalling frame was ranged from 1 to 7 using the exponential backoffalgorithm and from 0 to 6 using the splitting tree algorithm. When the exponentialbackoff algorithm was used it was found that by allocating at least 3 or 4 CSs persignalling frame, not only lower access delays were yielded but also an increase insystem throughput was achieved. Larger values (e.g. 5, 6 and 7 CSs) caused a waste ofbandwidth because more CSs are allocated than the number of CSs required to resolvecollisions. Similarly, a low system performance is also obtained when short values (e.g.1 and 2 CSs) are considered. Here the system inefficiency is attributed to an increasednumber of collisions, due to a reduction of slots for contention access. Results for thesplitting tree algorithm revealed that optimal system performance is achieved byallocating at least 1 CS per signalling frame. The increase in performance is achievedbecause an extra slot is reserved in the next signalling frame when a collision happens.This extra slot (is split into three minislots that carry shortened reservation requests) andis used only by the stations that caused the collision, so that new arrivals do not competewith backlogged stations for contention access. The increase in system throughput isachieved because by allocating at lest 1 CSs in each signalling frame, more slots can bescheduled for data transmissions.8-8

Chapter 8Final conclusions pWhen the performance of the Simple-CSA, Forced-CSA and Variable-CSA wascompared, it was found that the Forced-CSA outperformed the other two mechanisms.Table 8.2 presents a summary of the maximum system performance achieved by eachCSA for three different traffic configurations. The Forced-CSA not only provided thelowest access delays, but also (in most of the scenarios) the highest system throughputfor the exponential backoff algorithm, as appreciated in Table 8.2. The increase inperformance was because this mechanism allocates more contention slots when they areneeded to resolve collisions. By allocating two additional contentions slots in thefollowing signalling frame, after a collision has been detected, the probability of a newcollision among backlogged stations and new arrivals is decreased. This in turn, resultsin a reduced Contention-Resolution-Grant Cycle that cannot be obtained with the otherstrategies.This chapter also presented a performance optimisation and a comparison between theexponential backoff algorithm and the splitting tree algorithm. The first analysis studiedTable 8.2- Summary of maximum system performance for different CSAs.TrafficTypeInternet32 kbpsInternet64 kbpsMixed41.7kbpsCSA Min. CSs persign. frameMaximumThroughput (%)Mean AccessDelay (ms)CSs perRequestActiveStationsSimple3 46.8 3099.98 3.96344 51Simple5 48.0 2004.82 3.57891 51Forced-FSs 1 3 51.1 1284.42 2.87358 51Forced-FSs 2 2 51.5 584.026 2.81989 51Forced-FSs 2 3 51.0 1090.09 2.91893 51Variable2 47.9 2628.13 3.68146 51Variable3 47.8 2195.96 3.6845 51Simple 4 50.6 1226.38 3.07588 25Forced-FSs 1 3 50.7 913.49 3.01059 25Forced-FSs 2 2 50.6 494.56 3.01297 25Forced-FSs 2 3 49.8 416.05 3.15268 25Variable 2 50.8 777.72 2.95227 25Variable 3 51.3 624.79 2.87162 25Simple 3 43.9 925.94 3.18 34Simple 5 45.23 178.62 2.88 34Forced-FSs 2 2 45.6 417.15 2.84 34Forced-FSs 2 3 45.6 162.18 2.82 34Variable 2 44.2 1061.55 3.09 34Variable 3 44.6 309.13 3.01 34The shaded rows present the CSA with optimum system performance.8-9

Chapter 8Final conclusions pfocused on the optimisation of the initial and truncated backoff bounds (or backoffwindow Bw[i-t]) for the exponential backoff algorithm. Results for mixed traffic at 41.7kbps per station revealed that an increase in system performance is achieved with Bw[4-6]. With this backoff window, up to 75% of all data packets are transmitted in less than100 ms.A good system performance can also be obtained with Bw[3-4], Bw[2-4], Bw[3-6], andBw[3-5]. From this analysis it was seen that defining short values for the initial backoffexponent, such as Bw[2-3], results in a poor system performance. This is becausebacklogged stations are forced to transmit in the next 2 i = 4 contention slots, whichincreases considerably the risk of collision with new incoming packets. A similarperformance degradation is obtained with a large backoff window (e.g. Bw[5-7]). Theconsequence of this is because backlogged stations are now forced to wait a relativelylong period of time before they can compete again for contention access. With Bw[2-3]and Bw[5-7] only up to ≈ 33% of all packets generated are transmitted in less that 100ms.We also found the backoff window that provided optimal performance when only VoIPtraffic (9.7 kbps) is transmitted. They were Bw[5-7] and Bw[4-8]. If we compare therange of the optimum backoff window for mixed traffic and VoIP streams, we can seethat larger backoff windows are needed when only VoIP traffic is transmitted. This is tobe expected since the data rate of VoIP stream is lower than the data rate for mixedtraffic, thus the number of streams (or stations) can be increased. This in turn increasesthe probability of collisions, and indicates that larger backoff windows should beconsidered to reduce this probability. An important point here is that the maximumsystem throughput is reduced from ≈ 45% when mixed traffic is transmitted to ≈ 32%when only VoIP is delivered. This reduction is attributed to the increased number ofstreams supported, which requires more bandwidth to resolve collisions, and the higherprotocol overhead involved when delivering VoIP traffic.Similarly, the second performance optimisation focused on the Entry spreading (Es)factor of the splitting tree algorithm. In this analysis it was found that there is not muchdifference from selecting different Es values. Simulations results suggested that for8-10

Chapter 8Final conclusions poptimum system performance the Es factor should be set to 6 for mixed traffic (ormedium size networks) and 6 or 7 for VoIP traffic (or large networks). In terms ofsystem throughput, the difference of using distinct Es values was much less significant.It was found that the maximum system throughput ranged between ≈ 54.5% to 55% formixed traffic and from 44.8% to 45.2% of the cc for VoIP traffic.The last analysis presented in Chapter 6 approached a performance comparison betweenboth CSAs. In this analysis it was concluded that the splitting tree algorithmoutperforms the exponential backoff algorithm. This is because the former usesfeedback and allocation information that allows a station, with new incoming arrivals, tocompete for contention slots without the risk of collision with backlogged stations. Themost important advantage is that the use of minislots for reservation requests furtherdecreases the risk of collisions, since one contention-based slot is divided into threeminislots, increasing the probability of successful request transmissions. The majordrawbacks of this algorithm are higher complexity at the headend, increased processingtimes of the feedback and allocation information at the station, and since everycontention slot should be acknowledged regardless of whether it is used or not, highercontrol information at the downstream channel is assigned. In general, simulationresults showed that an increase over 9% on system performance, could be achieved withthe splitting tree algorithm when the backoff values (initial/truncated) and the Entryspreading factor have been optimised, for different traffic configurations (such as:Internet, VoIP and mixed traffic).In Chapter 7 several novel improvements were implemented. They enable theDVB/DAVIC MAC protocol to provide the delay requirements and an increasedefficiency for the support of timing critical interactive services and high-speed datatransmissions. The first analysis focused on the performance increase achieved by theuse of six enhanced-reservation-request mechanisms (Reserved Request, ContinuousReserved Request, Enhanced-Pure Reservation Access, Piggyback Request, ContinuousPiggyback Request and Unsolicited Grant Slot).The introduction of these techniques was because the DVB/DAVIC protocol uses alimited reservation access mechanism (PRA) by default for the transmission of8-11

Chapter 8Final conclusions preservation requests that was not optimised for the delivery of isochronous streams(timing critical interactive services), and this is evident from the increased risk ofcollision with reservation requests among stations and the poor system performance asseen in previous analysis. In general, it was demonstrated that by adopting theseenhanced mechanisms not only lower access delays are possible but also the systemthroughput is considerably improved. The highest system performance is produced withthe UGS mechanism. Results for VoIP traffic revealed that the number of streamssupported could be greatly increased from 105 to approximately 170 connections,achieving also relatively low access delays in the order of 4-7 ms. This appeared to bethe highest performance increase for this research, using only the contention and thereservation access modes of the DVB/DAVIC protocol. The CPG mechanisms alsooffered a good system performance. With this technique, the number of supported VoIPstreams can be increased up to ≈ 125.From this analysis, it was highlighted that the use of minislots with or withoutpiggybacking requests, and the enhanced pure reservation-access mechanism are bestsuited for Internet traffic or bursty traffic.In general, the six enhanced mechanisms achieve an increased performance over thedefault PRA, because as they become more complex, they efficiently reduce and insome cases avoid the collision risk of reservation request transmissions and thereforethe CRGC is minimised. The increase in system throughput is achieved because ascollisions are avoided, these mechanisms dynamically allocate more bandwidth for datatransmissions. An important point here is that the performance increase provided bythese mechanisms can only be obtained when the traffic characteristics are carefullyanalysed. The choice of a poor mechanism may reduce the service quality considerably.The performance of the piggyback mechanisms and the CRR mechanisms is alsoreported in [101], where a comprehensive analysis in terms of the cumulativeprobability of packet delay was presented. Due to their enhanced performance, thesethree mechanisms have already been integrated into the latest version of theDVB/DAVIC standard ‘ETSI EN 200 800’ [35].8-12

Chapter 8Final conclusions pA second approach for increased performance was through the use of Quality ofService. We have found that the main drawback of using the reservation access modefor the provision of delay sensitive applications is that bandwidth cannot be guaranteed.This is because all traffic transmitted using this mode is treated with a ‘best-effort’service and in some situations the delivery of TCIS requires an especial treatment for animproved service.Therefore, in order to provide a guaranteed service for TCIS the use of QoS isfundamental. Two types of QoS were studied. They are known as prioritisation andreservation.Although traffic prioritisation is not part of the set of functionalities of theDVB/DAVIC protocol specification, we have demonstrated that a faster transmissionfor the delivery of TCIS streams can be achieved by mapping the ToS field of the IPprotocol, with at least two 2-levels of priority at the DVB/DAVIC MAC layer. This willgive a faster delay treatment to TCIS stream. Results from a mixed traffic configuration(at 41.7 kbps) revealed that after the saturation point, the prioritised mechanism stilloffers tolerable delays (about 50 ms) for TCIS streams. Here, the number of VoIPstreams was increased from ≈ 40 with the default PRA up to ≈ 50 connections with theprioritised mechanism. The major drawback of this technique is that it does not provideany guarantee of bandwidth availability or latency with high traffic loads.On the other hand, the reservation technique uses the fixed access mode of theDVB/DAVIC protocol to provide a guaranteed service. Here the number of slots neededby a station and the periodic intervals is negotiated during the connection setup. Withthis technique, we have seen that reduced packet access delays for the delivery ofisochronous streams, in the range of 1 and 2 ms could be guaranteed by theDVB/DAVIC protocol, before large periods of congestion are experienced. It was alsofound that for the same traffic configuration (at 41.7 kbps), on high traffic loads, theservice of IP traffic gets starved at the headend and the number of VoIP streams couldbe significantly increased up to 320. In order to avoid the starving service for IP traffic,the scheduler algorithm at the headend could be slightly modified to control the8-13

Chapter 8Final conclusions pbandwidth to be assigned to each access mode (e.g. 40% of the cc for the fixed-rateaccess and 60% for the reservation and the contention access mode).The best performance of the reservation technique was achieved when only TCSIstreams were transmitted. Here we studied the performance of two codecs for VoIPtraffic (G.711 and G.723.1). Codec G.711 was considered to stress the CATV networkand mainly because EuroModems (class B) are more likely to use it for high qualityvoice calls. In addition, header suppression, which is an advanced functionality in thelatest version of this protocol [35], was also considered in this analysis. In general, itwas demonstrated that the use of header suppression, combined with a larger voiceframe (30 ms for G.711 and 120 ms for G.723.1), bandwidth efficiency is considerablyincreased to a large extent, achieving a much higher figure regarding the maximumnumber of streams sustainable. In addition, in this research a minimum packet accessdelay about 1ms (for VoIP streams) was possible. However, with advancedsynchronisation techniques this delay can be further reduced (under 1ms), for thedelivery of TCIS streams in general.For example, results for codec G.711 revealed that the number of streams supportedcould be augmented up to 53 connections when header suppression is considered andthree audio frames are encapsulated in each audio packet, compared to 27 connectionsachieved when only voice packets consist of one voice frame and header suppression isavoided. This corresponds to an increase of ≈ 96%. Similarly, for codec for G.723.1,this increase, in terms of streams supported, could be up to 400%. In order to achievethese remarkable figures, we have assumed that the RTP, UDP, IP and MAC headerscan all be suppressed, which is reasonable if the INA maintains additional stateinformation on all active voice connections in accordance to [47].Finally, the last analysis provided in this dissertation addressed a comparison betweenthe DVB/DAVIC and the DOCSIS protocol. We have evaluated some of thefundamental properties provided by each standard at the MAC and PHY layers. Ingeneral, results revealed that in most of the analysis, the DVB/DAVIC protocolachieved a reduced performance compared with the DOCSIS protocol. This was mainlybecause the DVB/DAVIC protocol encapsulates every datagram into ATM-AAL5 cells8-14

Chapter 8Final conclusions pand suffers an overhead penalty for Segmentation and Reassembly (SAR) in an attemptto provide a faster transmission for QoS over the ATM protocol, while the DOCSISuses a scheme that favours the delivery of variable length Internet protocol packetsrather than ATM transfer.8.3 Application of results to the IndustryThe results presented in this dissertation can be used in implementation for theEuropean cable modem protocol (DVB/DAVIC) by cable network operators andmanufactures. For example, results for mean access delay, throughput and utilisation(discussed in Chapter 5) can act as a guideline as to how subscribers should be allocatedin the upstream in order to achieve a desired performance and service level balance.In addition, by using the given performance levels for different loading scenarios, anappropriate charging scheme can be selected. Analysis on capital investment may bebased on results presented herein by deriving how the network could be segmented inorder to provide services with specific performance requirements. In general,recommendations for operators and service providers on how to achieve optimumsystem performance are as follows.The effects of the ‘Maximum Contention Access Message Length’ (Section 5.4.3)revealed that an increase in the overall system performance could be up to 3% by settingthe this parameter to 1. This parameter can be modified by the operator using the devicemanager at the headend. The ‘Maximum Reservation Access Message Length’ can alsobe modified by the service operator. A recommendation is to set this parameter to 32slots if the subscriber service is only for Internet traffic. If the service is also for voiceover IP, (with a best effort service) this parameter should be reduced to about 6 slots,but there is a drawback, reducing this parameter increases the risk of collisions ofreservation request and this results in a loss in system performance of about 6% in anattempt to provide short delays for the delivery of VoIP streams (as indicated in Figure5.12), on high periods of congestion.8-15

Chapter 8Final conclusions pThe optimisation of the number of contention slots carried out in Sections 6.2.2.1 and6.2.2.2 for the exponential backoff algorithm and the splitting tree algorithm,respectively, showed that a simple contention slots allocator can increase performanceby up to ≈ 10% (Figure 6.2 and Figure 6.5). This mechanism can be directlyimplemented by manufacturers by slightly modifying the DVB/DAVIC MAC protocol.The only parameter that is needed for the operation of the Simple-CSA is the ‘Min. o.of CSs per signalling frame’ , which was not defined by the DVB/DAVIC standard.This parameter can be added to the set of operational parameters of the INA and thenmodified via the device manager, either by the manufacturer or by the operator. Resultspresented in Section 6.2.2.1 can be used as a guidance of how to set the correct numberof contention slots (for optimum performance).Similarly, the Forced-CSA can also be implemented by the manufacturer by simpleadding the parameter forced contention slots (FSc), which represents the number ofslots added to the signalling frame after a collision occurred. This scheme was the bestcontention slot allocator, because it dynamically adjusts the number of contention slotsneeded to resolve collisions according to the traffic load, and achieves an increase in theoverall system performance by up to 3% more than the figure reported by the Simple-CSA. We recommend setting this parameter to 2 for optimum system performance.The optimisation of the two contention resolution algorithms used by the DVB/DAVICprotocol (approached in Section 6.3) does not require the intervention of themanufacturer. For the exponential backoff algorithm we recommend to the operator touse the backoff windows of Bw[4-6] and Bw[4-8], which performs well for Internet andVoIP traffic. For the splitting tree algorithm the Entry spreading factor should be set to6 for best system performance.From results presented in Section 6.3.3, we recommend that cable network operatorsbuy EuroModems that support the splitting tree algorithm. In this Section 6.3.3 it wasshown that this algorithm always performs better than the exponential backoffalgorithm, regardless of the traffic load and number of active stations. AlthoughEuroModems with this functionality tend to be more expensive than EuroModemssupporting only the exponential backoff algorithm, this extra cost is compensated for by8-16

Chapter 8Final conclusions pthe increase in system performance, which can be of up to ≈12% of the maximumchannel capacity.There are however modifications proposed herein, namely modifications at the INAscheduler and the NIU for the support of enhanced reservation request mechanisms. Thepiggyback mechanism and the continuous piggyback have already being integrated inthe third version of the DVB/DAVIC protocol [35] due to their remarkably performanceover the default reservation mechanism (defined in the second version). We stronglyrecommend to cable network operators to buy EuroModems supporting these piggybackmechanisms, here it was proved that the increase in system throughput could be up to25% of the maximum channel capacity for upstream channels supporting Internet andvoice traffic (see Section 7.2.7, Figure 7.8).The unsolicited grant slots (UGS) mechanism, however can achieve a better systemperformance than the piggyback mechanisms in terms of access delays for the supportof CBR traffic, here access delays below 7 ms can be obtained on high traffic loads. Theimplementation of this mechanism requires slight modifications to the INA schedulerand NIU MAC protocol. Here the ‘IA_Capabilities’ field of the Default ConfigurationMessage and the ‘IU_Capabilities’ field of the Sign-on Response Message, contain areserved field (of 5 bits for the) for future implementations, which can be used for thesupport of the UGS mechanism. This functionality requires only one bit, leaving theother four bits for future implementation. This reserved field is of one bit in the lastversion of the DVB/DAVIC standard [35], which can also be used for this functionality.With the UGS mechanism, as stated in Section 7.2.6, the number of slots needed by astation and the periodic intervals can be negotiated during the connection set-up thoughthe use of the ‘Resource Request Message’. This structure is the same as the one usedfor the establishment of fixed-rate connections, with the exception that instead ofassigning slots to a NIU in the fixed-rate region of the signalling frame, these slots areallocated in the reserved access region of the signalling frame, using the ‘ReservationGrant Message’.8-17

Chapter 8Final conclusions p8.4 Future workIn this dissertation we have provided a rigorous performance analysis and introducednovel techniques that increase the network efficiency of the upstream channel.However, there are many interesting issues requiring further research.The latest version of the DVB/DAVIC protocol specification (v.3) [35] has beenrecently released. New features such as header suppression, piggyback mechanisms,connection priority and the performance over a 12 Mbps upstream channel need to beresearched.In this dissertation we have seen that the use of header suppression provides the highestnetwork efficiency. In the latest version, a different approach is used for the suppressionof RTP sessions, which consists of the combination of RTP/UDP/IP headers. Instead ofmaintaining additional state information on all active connections at the headend, so thatthese headers and also the MAC header can be all suppressed (as suggested in [47]),only the fixed fields of the RTP/UDP/IP headers are suppressed in [35], as illustrated inFigure 8.1. This means that only up to 27 bytes (out of 12-bytes RTP + 8-bytes UDP +20-bytes IP = 40 bytes) can be suppressed. An interesting issue here would be to analysethe change in system performance when this strategy is used for the delivery of TCIS.8-18

Chapter 8Final conclusions pThe continuous piggyback request and the piggyback request mechanisms have alreadybeen included as part of the functionality of the DVB/DAVIC MAC protocol.The performance comparison of the enhanced-reservation request mechanisms providedin this dissertation considered fixed piggyback values (e.g. 0, 4, 8 and 12 for PG and 1,4, 8 and 12 for CPG). According to the new protocol specification these values (withthe exception of the first one, which is always fixed) could be up to 255. An additionalanalysis would be to provide a performance optimisation of the piggyback values forthese two strategies using different traffic configurations (e.g. Internet traffic, burstytraffic, VoIP, Video conferencing, multimedia traffic, etc.).The analysis of connection priorities would allow the DVB/DAVIC protocol toprioritise traffic transmissions. A connection with low priority can be reprovisioned inorder to accommodate the requirements of connections with high priority. The newprotocol specification defines up to 255 levels of priorities as indicated in Table 8.3.This functionality would require an improved scheduler algorithm at the INA, whichgives preference to connections with higher priority.Table 8.3 – Priorities for connections.ApplicationPriority valuesStandard data flow applications 0-7932 - bitIP headerUDP headerRTP headerversion IHLToSlengthID numberflagsfragmentation offsetTTL protocol header checksumsource addressdestination addresssource portdestination portlengthheader checksumv P X CC MPTsequencetime stampSSRCCSRC listIHL: Internet Header LengthToS: Type of ServiceTTL: Time To Livev: versionP: PaddingX: Extension bitCC: CSRC CountCSRC: Contributing source identifiers listSSRC: Synchronisation sourcefixed fieldvariable fieldFigure 8.1 – Fixed fields for RTP/UDP/IP headers.8-19

Chapter 8Final conclusions pApplications with QoS 80-200High priority applications 201-255In this research, most of the analysis focused on 3 Mbps and 6 Mbps upstream channels.An interesting issue is to analyse the performance when applications that demand higherbandwidth requirements are delivered using a 12 Mbps channel.Another important topic that needs to be addressed as future work is quality of servicein the downstream direction. The mapping of up to eight upstream channels onto asingle downstream channel raises the issue of congestion in the downstream too. For thetransmission of traffic that does not originate in the CATV network, for example trafficcoming from either the Internet, PSTN or other CATV network via the router/ATMswitch attached to the INA, an efficient scheduler will be required in order to prioritiseand police traffic as well as maintain maximum link utilisation in the downstreamchannel.Finally, the work presented in this dissertation is highly relevant to emerging fixed,mobile and satellite architectures [20]. Fixed wireless architectures have the sameaccess topology as CATV networks. Figure 8.2 presents three different configurationsof CATV networks with satellite, terrestrial or HFC architectures. The spectrum isdivided into different unidirectional channels for the upstream and downstreamdirection. The major difference is the physical interface. In wireless architectures, thesignals propagate through the air while in CATV networks either coax or optic fibrelinks or a combination of both is used for transmissions. The physical interface wouldbe significantly different but the MAC would be identical.CATV networks have been considered as the ideal backbone network architecture forPersonal Communications Service (PCS), as illustrated in Figure 8.2c, [50]. Thebroadcast nature of CATV networks significantly simplifies the hand-off mechanism ofmobile stations thus reducing cost and complexity at both the base station and thetelecom switch.8-20

Chapter 8Final conclusions pThe major drawback of using CATV networks for satellite communications, as seen inFigure 8.2a, is the long propagation link delays, which can result in serious performanceimplications.This is because very large propagation delays (in the order of hundreds of milliseconds)will results in equally large scheduler look-ahead delays. The propagation delay to theheadend and back in such systems will be twice the roundtrip delay to the satellite,which in the case of Geo-stationary Earth Orbit (GEO) satellites placed at about 22.3miles above the Earth would be of ≈ 500 ms [9]. Therefore the explicit CRGC cycle willresult in large intervals of network idle time due to the number of contention slots the schedulerwill have to subscribe.For delay sensitive applications, such as voice conversations, the propagation delay overthe two-way path between conversants influences the perceived quality of service. Thisproblem can be ameliorated with LEO satellites placed at ≈ 500 – 1000 miles above theEarth. This relatively short distance reduces transmissions delays significantly to ≈ 50ms [9]. The drawback is that a large number of satellites, arrayed on a constellation ofmultiple satellites in a polar orbit is required to maintain path continuity.The long propagation delays of satellite systems undoubtedly will reduce the systemefficiency of a CATV MAC protocol. An interesting issue requiring further research isto analyse to what extent long propagation delay can degrade the quality of service ofGEOBACKBONECity xPSTNCity yLEOFDDIDQDBMobile NodeISPMobile NodeHeadendHeadendMobile NodeMobile Nodea) Satellite Mobile/Fixed b) Digital CATV b) Terrestrial Mobile/ Fixed WirelessFigure 8.2 – Similarities of CATV with fixed/mobile terrestrial and satellite wirelessarchitecture.8-21

Chapter 8Final conclusions pTCIS streams over the DVB/DAVIC protocol. The direct consequence of having longpropagation delays is that the contention resolution algorithm interval will increasesimply because each station will have to assume longer time-out periods for thereception of a grant. In the case where a collision does not occur this will result in theCRA to be initiated with a delay equal to two times the round trip. If the traffic patternof a station is of such a nature (e.g. sporadic packet generation), this results in the use ofcontention access for the transmission of reservation requests, and performance can bedegraded significantly.8-22

REFERENCES[1] N. Abramson, “Development of the ALOHANET”, IEEE Trans. on Info. Theory, Vol. IT-31, No.2, Mar. 1985.[2] N. Achilleoudis, “DOCSIS 1.0 Key Achievements”, Cisco Systems, Inc. May 3, 1999.[3] J. Aldous, “Ultimate instability of exponential backoff protocols for acknowledgementbased transmission control of random access communications channels”, IEEE Trans. onInfo. Theory, Vol. IT33, pp. 219-223, 1987.[4] A. O. Allen, “Introduction to Computer Performance Analysis with Mathematica”, APProfessional, ISBN 0-12-051070-7, pp 9-39, 1994.[5] ATM Forum, “ATM Forum Technical Committee: Residential Broadband ArchitecturalFramework”, AF-RBB-0099.000, ATM Forum, Jul.1998, http://www.atmforum.com/[6] A. Azzam, “High Speed Cable Modems”, McGraw-Hill, ISBN 0-07-006417-2, 1997.[7] J. Bajon and G. Fontaine, “Development of Digital Television in the European Union”,Institut de L’audiovisuel et des Télécommunications en Europe (IDATE), Reference report2000, Jun. 2001[8] H. Barton, “White Paper: DOCSIS MCNS vs. DVB/DAVIC DVB-RCC”, MD,Broadcentric Ltd., May 1999.[9] R. J. B. Bates, “Broadband Telecommunications Handbook”, McGraw-Hill,Telecommunications, ISBN 0-07-134648-1, 2000.[10] D. Bertsekas and R. Gallager, “Data Networks”, Prentice-Hall, 2 nd Ed., Englewood Cliffs,NJ, ISBN 0-13-200916-1, pp. 271-362, 1992.[11] V. K. Bhagavath, “Emerging High-Speed xDSL Access Services: Architectures, Issues,Insights, and Implications”, IEEE Commun. Mag., Vol. 37, No. 11, pp.106-114, Nov.1999.[12] C. Bisdikian, K. Maruyama, D. I. Seidman, and D. N. Serpanos, “Cable Access Beyond theHype: On Residential Broadband Data Service over HFC Networks”, IEEE Commun.Mag., pp. 128-135, Nov. 1996.[13] Cable Modem Report, “Despite Service Provider Pratfalls, Cable Modem SubscriberGrowth Remains Robust”, Multimedia Broadband Services & Infrastructure, CableIndustry. Report No. MB01-15DC, Dec. 2001.http://www.instat.com/abstracts/DT/2001/MB0115DC_abs.htm[14] J. I. Capetanakis, “Tree Algorithm for Packet Broadcasting Channel”, IEEE Trans. on Info.Theory, Vol. IT-25, pp. 505-515, Sep. 1979.[15] S. H. Cho, J. H. Kim and S. H. Park, “Performance Evaluation of the DOCSIS 1.1 MACProtocol According to the Structure of a MAP Message”, ICC-2001, ISBN 0-7803-7097-1,Helsinki, Jun. 2001.[16] M. Christoph, “Cable Networks in Europe - an Overview of Today’s and Tomorrow’sTechnologies”, Proc. of the IBC-98, pp. 44-48, Sep. 1998.R-i

Referencesp[17] I. Cidon, and M. Sidi, “Conflict Multiplicity Estimation and Batch Resolution Algorithms”,IEEE Trans. on Info. Theory, Vol. IT-34, No. 1, pp. 101-110, Jan. 1988.[18] R. Citta, C. C. Lee, and D. Lin, “Phase 2 Simulation Results for Adaptive Random AccessProtocol”, IEEE802. 14 WG, Contribution No. IEEE802.14-96/114, May 1996.[19] P.W. Copper and J. F. Bradley, “A satellite network for Internet access in space”, IEEEspectrum, pp. 32-33, Jan. 1999.[20] M. Corner, N. Golmie, J. Liebeherr and D. Su, “A Priority Scheme for the IEEE 802.14MAC Protocol for Hybrid Fiber-Coax Networks”, Proc. of INFOCOM'98, ISSN 1063-6692, pp. 1400-1407, San Francisco, CA, Mar. 1998.[21] DOCSIS 1.0, “Data-Over-Cable Interface Specifications - Radio Frequency InterfaceSpecification”, CableLabs, SP-RFI-I04-980724, Jul. 1998.[22] DOCSIS 1.1. “Data-Over-Cable Interface Specifications - Radio Frequency InterfaceSpecification”, CableLabs, SP-RFIv1.1-I01-990311, Mar. 1998.[23] DOCSIS 1.2. “Data-Over-Cable Interface Specifications - Radio Frequency InterfaceSpecification”, CableLabs, SP-RFIv2.0-I01-011231, Dec. 2001.[24] G. Donaldson and D. Jones, “Cable Television Broadband Network Architectures,” IEEECommuns. Mag., Vol. 39, No. 6, pp 122-1226, Jun. 2001.[25] C. Doyle, “The Dynamics of Local Access: Telecommunications in New Zealand”, LondonEconomics, Apr. 2000.[26] C. Drewes, W. Aicher and J. Hausner, “The wireless art and the wired force of subscriberaccess”, IEEE Communs Mag., Vol. 39, No. 5 , pp. 118 –124, May 2001.[27] C. A. Eldering, N. Himayat, and F. M. Gardner, “CATV Return Path Characterization forReliable Communications”, IEEE Communs. Mag., Vol. 33, No. 8, pp. 62-69, Aug. 1995.[28] H. ElGebaly, “Characterization of Multimedia Streams of an H.323 Terminal”, IntelTechnology Journal Q2, 1998.[29] EN 300 421 v1.1.2, “Framing structure, channel coding and modulation for 11/12 GHzsatellite services”, ETSI, Aug. 1997.[30] ESIS Report, “Information Society indicators in the Member States of the EuropeanUnion”, A report prepared by the ESIS Project Management Support Team, InformationSociety Activity Centre, European Survey of information Society (ESIS), Oct. 2000,(http://europa.eu.int/ISPO/esis/default.htm)[31] ETS 300 429 v1.2.1, “Digital Video broadcasting: Framing structure, channel coding andmodulation for cable systems”, ETSI, Apr. 1998.[32] ETS 300 800, “Digital Video Broadcasting: Interaction Channel for Cable TV DistributionSystems (CATV)”, ETSI, Jul. 1998.[33] ETSI EN 301 192 v1.2.1, “Digital Video Broadcasting (DVB); DVB specification for databroadcasting”, ETSI, Jun.1999.[34] ETSI ES 200 800 v.1.2.1, “Digital Video Broadcasting: Interaction Channel for Cable TVDistribution Systems (CATV)”, ETSI, Apr. 2000.R-ii

Referencesp[35] ETSI ES 200 800 v.1.3.1, “Digital Video Broadcasting: Interaction Channel for Cable TVDistribution Systems (CATV)”, ETSI, Oct. 2001.[36] EuroModem, “Technical Specification of a European Cable Modem for digital bidirectionalcommunications via cable networks”, ECCA & EuroCableLabs, v.1, May 1999.[37] M. Gagnaire, “An Overview of Broadband Access Technologies”, Proc. of the IEEE, Vol.85, No. 12, pp. 1958-1972, Dec.1997.[38] M. Draoli, M. Lancia and A. Laureti-Palma, “Video conferencing on a LAN/MANinterconnected system: QoS evaluation”, Proc. of the Fourth International Conference onComputer Communications and etworks, pp 170-177, 1995.[39] L. A. Goldberg and P. D. MacKenzie, M. Paterson and A. Srinivasan, “ContentionResolution with Constant Expected Delay”, Journal of the ACM, Vol. 47, No. 6, pp. 1048-1096, 2000.[40] N. Golmie, S. Masson, G. Pieris, and D. Su, “A MAC Protocol for HFC Networks: DesignIssues and Performance Evaluations”, Computer Communs, Vol. 20, pp. 1042-1050, 1997.[41] N. Golmie, Y. Saintillan, and D. Su, “A Review of Contention Resolution Algorithms forIEEE 802.14 Networks”, IEEE Communication Surveys, 1Q99, Vol. 2, no. 1, 1999. Alsoappears in a book entitled, “Cable Modems: Current Technologies and Applications”,IEEE Press, ISBN 0-7803-5395-X, pp. 233-260, 1999.[42] N. Golmie, F. Mouveaux, and D. Su, “A Comparison of MAC Protocols for HybridFiber/Coax Networks: IEEE 802.14 vs. MCNS”, Proc. ICC'99, ISBN 0-7803-5284-X, Jun.1999.[43] N. Golmie, F. Mouveaux, and D. Su, “Differentiated Services over Cable Networks”, Proc.GLOBECOM'99, ISBN 0-7803-5796-5, Rio de Janeiro, Brazil, Dec. 1999.[44] A. G. Greenberg, P. Flajolet and R. E. Ladner, “Estimating the multiplicities of conflicts tospeed their resolution in multiple access channels”, Journal of the ACM, Vol. 34, No. 2,pp.289-325, Apr. 1987.[45] D. Gross, C. Harris, “Fundamentals of Queuing Theory”, Wiley Series in Probability andStatistics, John Wiley and Sons, Inc., 3 rd edition, ISBN 0-471-17083-6, 1998.[46] R. Gusella, “A Measurement Study of Diskless Workstation Traffic on an Ethernet”, IEEETran. on Communs., Vol. 38, No. 9, ISSN 0090-6778, pp. 1557-1568, Sep. 1990.[47] D. Hartman and T. J. Quigley, “Multimedia Cable-Network System Media Access-ControlPerformance Simulation”, Fijoleck, Kuska, Majeti, and Sriram (eds.), IEEE Press, CableModems: Current Technologies and Applications, Chicago, ISBN 0-7803-5395-X, pp 261-280, 1999.[48] J. Håstad, T. Leighton and B. Rogoff, “Analysis of Backoff Protocols for Multiple AccessChannels”, SIAM Journal on Computing, Vol. 25, pp. 740-774, 1996.[49] J. Heinanen: “Multi-protocol Encapsulation over ATM Adaptation Layer 5”, IETF RFC1483, Jul. 1993.R-iii

Referencesp[50] N. F. Huang, C. A. Su and H. C. Chao, “Architectures and Handoff Schemes for CATVbasedPersonal Communications Network”, IEEE IFOCOMM, ISBN 0-7803-4383-2,1998.[51] IEEE 802.14, “Cable-TV access method and physical layer specification”, IEEE 802.14Sub-committee, IEEE 802 Publications, Draft 2, Revision 1, Jul. 1997[52] IEEE 802.14, “Cable-TV access method and physical layer specification”, IEEE 802.14Sub-committee, IEEE 802 Publications, Draft 3, Revision 3, Oct. 1998.[53] IEEE 802.14, “Evaluation Models for IEEE 802.14 MAC Protocol”, IEEE 802.14 Subcommittee,IEEE 802 Publications, Ref. No. IEEE802.14/95-061R2. Jan.1996.[54] IEEE 802.14 HI-PHY, “FA-TDMA/S-CDMA Upstream HI-PHY Proposal”, Draft 1,Revision 2, IEEE 802.14 Sub-committee, IEEE 802 Publications, Mar. 1999.[55] Interoperability Consortium, “White paper: Overview of DVB-RCCL/DAVIC vs.MCNS/DOCSIS”, DVB/DAVIC Interoperability Consortium, Oct. 1998.[56] ITU-T Rec. G.711, “Pulse Code Modulation (PCM) of voice frequencies”, ITU-T, Nov.1988.[57] ITU-T Rec. G.723.1, “Speech coders: Dual rate speech coder for multimediacommunications transmitting at 5.3 and 6.3 Kbit/s”, ITU-T, 1996.[58] ITU-T Rec. G.992.2, “Splitterless asymmetric digital subscriber line (ADSL) transceivers”,ITU-T, Jul. 1999.[59] ITU-T Rec. H.323, “Packet-based Multimedia Communications Systems”, ITU-T, Geneva,Switzerland, Nov. 2000.[60] M. Ivanovich and M. Zukerman, “Evaluation of Priority and Scheduling Schemes for anIEEE 802.14 MAC Protocol Loaded by Real Traffic”, IFOCOM’98, ISBN 0-7803-4383-2, pp. 1384-1391, 1998.[61] P. Jacquet, P. Muhlethaler, and P. Robert, “Asymptotic average access delay analysis:adaptive p-persistence versus tree algorithm”, IEEE 802.14 Working Group, paper 96-248,1996.[62] P. Kelly and I. M. MacPhee. “The Number of Packets Transmitted by Collision DetectRandom Access Schemes”. Annals. of Prob., Vol. 15, No. 4. pp. 1557-1568, Oct. 1987.[63] D. Knott, “CABLE INTERNET ACCESS: Protocols, Pricing, and Deployment”, Texas A& M University, ENTC 425, Dec. 1999.[64] T. Kos, B. Zovko-Cihlar, and S. Grgic, “New services over CATV network”EUROCON'01, Inter. Conference on Trends in Communs., Vol. 2, pp. 442-445, 2001.[65] P. R. Kumar, “A tutorial on Some New Methods for Performance Evaluation of QueuingNetworks”, IEEE J. Selected Areas in Communs., ISSN 0733-8716, pp. 970-980, Aug.1995.[66] P. R. Kumar and S. P. Meyn, “Stability of Queuing Networks and Scheduling Policies”,IEEE Tran.. on Automatic Control, pp. 251-260, Feb. 1995.R-iv

Referencesp[67] T. C. Kwok, “A vision for residential broadband services: ATM-to-the-home”. IEEEetwork, Vol. 9, No. 5, ISBN 0-7803-2448-X, pp. 14-28, Jul. 1995.[68] T. C. Kwok, “Residential broadband architecture over ADSL and G.Lite (C.992.2): PPPover ATM”, IEEE Communs. Mag., Vol. 37, No. 5, pp. 84 –89, May 1999.[69] M. E. Laubach, D. J. Farber and S. D. Dukes, “Delivering Internet Connections overCable”, Wiley Networking Council Series, ISBN 0-471-38950-1, 2001[70] A. Law, M. McComas, “Simulation software for communications networks: the state of theart”, IEEE Communs. Mag., Vol. 32, No. 3, pp. 44 -50, Mar. 1994.[71] W. E. Leland et al., “On the Self Similar Nature of Ethernet traffic”, Proc. of Sigcomm '93,ISSN 1063-6692, Association for Computer Machinery, Ithaca, N.Y., 1993.[72] J.O. Limb and D. Sala, “An Access Protocol to Support Multimedia Traffic over HybridFiber/Coax Systems”, Second International Workshop in Community etworking,Princeton, ISBN 0-7803-2756-X, pp. 35-40, Jun. 1995.[73] Y. D. Lin, C. Y. Huang and W. M. Yin, “Allocation and Scheduling Algorithms for IEEE802.14 and MCNS in Hybrid Fibber Coaxial Networks”, IEEE Tran. on Broadcasting,Vol.44, No.4, ISSN 0018-9316, pp. 427-435, 1998.[74] R. Lloyd-Evans, “Wide Area Network Performance and Optimization”, Addison Wesley,ISBN 0-201-4270-0, pp. 1-20, 1996.[75] Y. Machida, “Technologies for Local Access Fibering”, IEEE Communs. Mag., Vol. 32,No.2, pp. 64-73, Feb. 1994.[76] T. Miki, “Toward the Service Reach-Era”, IEEE Communs. Mag., Vol. 32, No.5, pp. 34-39, Feb 1994,[77] Narayanaswamy, “MCNS Design Specification”, MIL3 Inc., Jan. 1997.[78] K. M. Nichols and M. Laubach, “On Quality of Service in an ATM-based HFCArchitecture”, Proc. of ATM’96, San Francisco, Aug. 1996.[79] OPNET Modeller v6.0, “OPNET Modeller Simulation Package”, OPET Technologies(MIL 3), 1997, http://www.opnet.com[80] PacketCable, “PacketCable Audio/Video Codecs Specification”, CableLabs,PacketCable Project, Dec. 2001.[81] A. J. Phillips et al., “SuperPON: A Broadband Access Solution From ACTS-Planet”, 14 thTeletraffic Symposium, Apr. 1997.[82] QoS White Paper, “The Need for QoS”, QoS Forum, Jul. 1999. www.qosforum.com[83] P. Raghavan and E. Upfal, “Stochastic Contention Resolution With Short Delays”, SIAMJournal on Computing, Vol. 28, No. 2, pp. 709-719, 1998.[84] V. Rangel, C. Smythe, P. Tzerefos, S. Cvetkovic and S. Landeros, “A comparison of theDOCSIS, DVB/DAVIC and IEEE 802.14 Cable Modem Specifications”, Proc. of theInternational Conference on Telecommunications (ICT), Acapulco, May 2000.R-v

Referencesp[85] V. Rangel, R. Edwards, P. Tzerefos and K. Schunke, "DELIVERY OF LOW RATEISOCHRONOUS STREAMS OVER THE DIGITAL VIDEOBROADCASTING/DIGITAL AUDIO-VIDEO COUNCIL INTERNATIONAL CABLETELEVISION PROTOCOL", Submitted to IEEE Tran. on Broadcasting, Nov. 2000.[86] V. Rangel, R. Edwards, and K. Schunke, “Contention Resolution Algorithms for CATVNetworks Based on the DVB/DAVIC Cable Modem Protocol Specification (ETS EN 200800)”, Proc. of the International Broadcasting Conference (IBC), Amsterdam, Sep. 2001.This paper is to be published in the journal Cable Telecommunication Engineering (CTE),Sep. 2002.[87] V. Rangel and R. Edwards. “Performance Analysis and Optimisation of the Digital VideoBroadcasting/Digital Audio Visual Council Cable Modem Protocol for the Delivery ofIsochronous Streams”. Proc. of GLOBECOM-2001, IEEE, ISBN 0-7803-7206-9, Vol. 1,pp. 430-434, Nov. 2001.[88] U. Reimers, “Digital Video Broadcasting: The International Standard for DigitalTelevision”, Springer, 1 st edition, ISBN 3-540-60946-6, Jan. 2001.[89] M. Reiser and H. Kobayashi, “Queuing Networks with Multiple Closed Chains: Theoryand Computational Algorithms”, IBM Journal of Research and Development, pp. 285-294,1975.[90] M. Reiser, “A Queuing Analysis of Computer Communication Networks with WindowFlow Control”, IEEE Tran. on Communs, pp. 1199-1209, Aug. 1979.[91] M. Reiser, “Mean Value Analysis and Convolution Method for Queue-Dependent Serversin Closed Queuing Networks”, Performance Evaluation 2, pp. 9-18, 1981.[92] R. L. Rivest, “Network Control by Bayesian Broadcast”, Technical Report MIT/LCS/TM-287, MIT Lab for Computer Science, Cambridge, Massachusetts, 1985.[93] D. Sala, D. Hartman, and J.O. Limb, “Comparison of Algorithms for Station Registrationon Power-up in an HFC Network”, IEEE 802.14 WG, Contribution No. IEEE802.14/96-012, Boulder CO, Jan. 1996.[94] D. Sala, and J. O. Limb, “Scheduling Disciplines for HFC Systems: What can we learnfrom ATM Scheduling?”, Third International Workshop in Community etworking, ISBN0-7803-3304-7, Antwerpen, Belgium, pp. 13-18, May 1996.[95] D. Sala, J.O. Limb, and S. Khaunte, “Performance of Contention Resolution Algorithmsusing Continuous-Mode Operation”, IEEE 802.14 WG, Contribution No.IEEE802.14/IEEE97-048, Irvine CA, Mar. 1997.[96] D. Sala, J.O. Limb, and S. Khaunte, “Adaptive MAC Protocol for a Cable Modem”,Georgia Tech Technical Report GIT-CC-97-014, May 1997.[97] D. Sala, and J. O. Limb, “Comparison of Contention Resolution Algorithms for a CableModem MAC Protocol”, International Zurich Seminar on Broadband Communications,Zurich, Switzerland, Feb. 1998.R-vi

Referencesp[98] D. Sala, J. O. Limb, and S. U. Khaunte, “Adaptive Control Mechanism for Cable ModemMAC Protocols”, Proceedings of IFOCOM'98, ISBN 0-7803-4383-2, San Francisco,CA, Mar. 1998.[99] D. Sala, “Design and Evaluation of MAC Protocols for Hybrid Fiber/Coaxial Systems”,Ph. D. Dissertation, Georgia Institute of Technology, Mar. 1998.[100] K. D. Schunke, “Performance Evaluation of the DVB/DAVIC cable return channelsystem ETS 300 800”, Proc of the IBC’98, Amsterdam, pp. 58-63, Sep. 1998.[101] K. D. Schunke, “Performance Optimisation of the European Cable CommunicationStandard for Timing-Critical Interactive Services”, Proc. of the IBC’00, Amsterdam, pp.58-63, Sep. 2000.[102] V. Sdralia, C. Smythe, P. Tzerefos, and S. Cvetkovic, “Performance Characterization ofthe MCNS DOCSIS 1.0 CATV Protocol with Prioritized First Come First ServedScheduling”, IEEE Tran. on Broadcasting, ISSN 0018-9316, Vol. 45, No. 2, pp. 196-205,1999.[103] V. Sdralia, P. Tzerefos, C. Smythe and S. Cvetkovic, “Recovery of DOCSIS 1.0 CATVNetworks after system wide service disruption”, 2nd GEMISIS Technical Symposium onMultimedia-network-technology, Salford, UK, May 1999.[104] V. Sdralia, P. Tzerefos, S. Cvetkovic and C. Smythe, “RELIABILITY OF DOCSISCATV NETWORKS AFTER LARGE SCALE POWER FAILURES”, Proc. of the ICT,Acapulco, Mexico, May 2000.[105] V. Sdralia, “Optimized Recovery of DOCSIS Networks using Reserved PersistentRanging”, Proc. IEEE Globecom 2001, Vol. 1, ISBN 0-7803-7206-9, San Antonio,Texas, USA, pp. 415-419, Nov. 2001.[106] V. Sdralia, P. Tzerefos and C. Smythe, “Recovery Analysis of the DOCSIS Protocol afterService Disruption”, IEEE Tran. on Broadcasting, Vol. 47, No. 4, ISSN 0018-9316, pp.377-385, Dec. 2001.[107] D. Senie, “Network Address Translator (NAT)-Friendly Application Design Guidelines”,Internet - Draft: RFC 3235, Jan. 2002.[108] C. Smythe, P. Tzerefos, I. Stergiou and C. Sullivan, “A Comparison of the IEEE 802.14Broadband Metropolitan Area Network Protocol and the MCNS Cable ModemSpecification”, IBC Television Distribution Technology, pp. 52-57, Sep. 1998.[109] C. Smythe, P. Tzerefos and S. Cvetkovic, “CATV Infrastructures and Broadband DigitalData Communications”, Contribution on the Encyclopedia for of Electrical andElectronics Engineering, John Wiley & Sons, pp. 21, 1998.[110] C. Smythe et al., “Performance Evaluation of the DVB/DAVIC Cable Return ChannelPath for Interactive Services”, IBC DVB’99, pp. 1-23, Apr.1999. Twice[111] C. Smythe, P. Tzerefos, V. Sdralia and V. Rangel, “Cable Modems and the ReturnChannel Path for Interactive Services: DOCSIS Vs. DVB – A Performance Evaluation”,IBC Television Distribution Technology, pp. 1-28, May 1999.R-vii

Referencesp[112] K. Sriram, “Methodologies for Bandwidth Allocation Transmission Scheduling, andCongestion Avoidance in Broadband ATM Networks”, Computer etwork & ISDSystems, Vol. 26, pp. 43-59, 1993.[113] K. Sriram, “Performance of MAC protocols for Broadband HFC and Wireless AccessNetworks”, Advances in Performance Analysis, Vol. 1, No. 1, pp. 1-37, Mar. 1998.[114] W. Stallings, “Local and Metropolitan Area Networks”, Prentice Hall InternationalEditions, 5 th edition, ISBN 0-13-253733-8, pp. 426-460, 1997.[115] W. Stallings, “Self similarity upsets data traffic assumptions”, IEEE Spectrum, pp. 28-29.Jan.1997,[116] S. Tanenbaum, “Computer Networks”, Prentice-Hall, 3 rd edition, ISBN 0-13-394248-1,pp. 245-254, 1996.[117] F. A. Tobagi and V. B. Hunt, “Performance Analysis of Carrier Sense Multiple Accesswith Collision Detection”, Computer Networks 4, pp. 245-259, 1980.[118] TR 101 329 V2.1.1, “Telecommunications and Internet Protocol Harmonization OverNetworks (TIPHON); General Aspects of Quality of Service (QoS)”, ETSI, Jun. 1999.[119] S. Tsybakov and V. A. Mikhailov, “Free Synchronous Packet Access in a BroadcastChannel With Feedback”, Problems of Information Transmission, Vol. 14, No. 4, pp.259-280, Oct- Dec. 1978.[120] P. Tzerefos, “On the Performance and Scalability of digital upstream DOCSIS 1.0conformant CATV channels”, Department of Computer Science, The University OfSheffield, Ph. D. Thesis, Oct. 1999.[121] P. Tzerefos, C. Smythe, V. Sdralia, and S. Cvetkovic, “Delivery of Low Bit RateIsochronous Stream Over the DOCSIS .10 Cable Television Protocol”, IEEE Tran. onBroadcasting, Vol. 45, No. 2, ISSN 0018-9316, pp. 206-214, Jun. 1999.[122] Y. Wakui, “The Fiber-Optic Subscriber Network in Japan”, IEEE Communs. Mag., Vol.32, No. 2, pp 56-63, Feb. 1994.[123] W. Weippert, “The Evolution of the Access Network in Germany”, IEEE Communs.Mag., Vol. 32, No. 2, pp 50-55, Feb. 1994.[124] J. Wolf and N. Zee, “The Last Mile: Broadband and the Next Internet Revolution”,McGraw Hill, ISBN 0-07-136349-1, pp. 21-35, 2001.[125] R. Yassini, “The Cable Modem's Evolution and Revolution: A Historical View”,Fijoleck, Kuska, Majeti, and Sriram (eds.), Cable Modems: Current Technologies andApplications, ISBN 0-7803-5395-X, Chicago, pp 16-21, 1999.R-viii

ReferencespWebsites[126] “Cable Datacom News”, Cable Modem Info Center, Last updated: Apr. 5, 2002,http://www.cabledatacomnews.com/[127] “CableLabs”, Revolutionizing cable technology, Last updated: 2002,http://www.cablelabs.com/[128] A CableLabs White Paper, “The Cable Connection: The Role of Cable Television in theNational Information Infrastructure”, Last updated: 2002,htttp://www.cablelabs.com/about_cl/pubs/cableNII.html[129] “Digital Audio-Visual Council (DAVIC)”, Last updated: Nov. 25, 1999, This web sitewill remain in operation until the end of Sep. 2002, http://www.davic.org/,[130] “Digital Video Broadcasting (DVB) Project”, Last updated: April 5, 2002,http://www.dvb.org/latest.html[131] “European Cable Communications Association (ECCA)”, Last updated: 2002,http://www.ecca.be/[132] “European Telecommunications Standards Institute (ETSI)”, Last updated: Apr. 5, 2002,http://www.etsi.org/[133] “PacketCable Project”, CableLabs Technology, Last updated: 2002,http://www.packetcable.com/[134] “Society of Cable Telecommunications Engineers (SCTE)”, Last updated: Apr. 5, 2002,http://www.scte.org/[135] “Beijing Environment, Science and Technology Update”, Produced weekly by the ESTSection, U.S. Embassy Beijing, Last updated: Jan. 28, 2002, http://www.usembassychina.org.cn/english/sandt/estnews1201.htmR-ix

APPEDIX A: GLOSSARYAALABRADSLATMBCBEBIMBNABONeSCAPCATVCC or ccCBRCIRCMCMTSCPECPGCPRCRGCCRRCSACSRCCSFDAVICDDICDESDHCPDMTDOCSISDSLATM Adaptation LayerAvailable Bit RateAsymmetrical Digital Subscriber LoopAsynchronous Transfer ModeBroadcast ChannelBest Effort serviceBroadcast Interface ModuleBroadcast Network AdaptorBlock Oriented Networks SimulatorCarrierless Amplitude/Phase modulationCommunity Antenna TelevisionChannel CapacityConstant Bit RateCommitted Information RateCable ModemCable Modem Termination SystemCustomer Premise EquipmentContinuous Piggyback RequestCentralised Priority ReservationContention-Resolution-Grant CycleContinuous Reservation RequestContention Slot AllocatorContributing source identifier listCommon Simulation FrameworkDigital Audio Video CouncilDVB/DAVIC Interoperability ConsortiumData Encryption StandardDynamic Host Configuration ProtocolDiscrete Multi-ToneData Over Cable System Interface SpecificationDigital Subscriber LoopA-i

Appendix AGlossary pDVBECCAECLEMENEPRAESETSETSIFA-TDMAFECFIFOFSMFTPFTTCFTTHGBHDSLHEHFCHSIBICIEEEIETFIHLIIMINAIPISDNISPITUDigital Video BroadcastEuropean Cable Communications AssociationEuroCableLabsEuro ModemEuropean NormEnhanced Pure Reservation-AccessEuropean StandardEuropean Telecommunication StandardEuropean Telecommunication Standard InstituteFrequency Agile – Time Division Multiple Access DivisionForward Error CorrectionFirst In First OutFinite State MachineFile Transfer ProtocolFibre To The CurbFibre To The HomeGuard BandHigh data rate Digital Subscriber LineHeadendHybrid Fibre CoaxialHeader SuppressionIn BandInteraction ChannelsInstitute of Electrical and Electronic EngineersInternet Engineering Task ForceInternet Header LengthInteractive Interface ModuleInteractive Network AdaptorInternet ProtocolIntegrated Systems Digital NetworkInternet Service ProviderInternational Telecommunications UnionA-ii

Appendix AGlossary pkbpsLANLLCMACMANMAPMCIMCNSMPEGMPEG-TSMSOsMVANATNICNIUNTTNVODONUOOBOPNETOSIPCIPCSPCMPDFPDUPGPHYPONPOTSPPPPRAQAMKilobits per secondLocal Area NetworkLogical Link ControlMedia Access ControlMetropolitan Area NetworkMAC Management AccessMAC Control InformationMultimedia Cable Network SystemsMotion Pictures Experts GroupMotion Pictures Experts Group Transport StreamMultiple Service OperatorsMean Value AnalysisNetwork Address TranslatorNetwork Interface CardNetwork Interface UnitNippon Telegraph and TelephoneNear Video On DemandOptical Network UnitOut Off BandOPtimised Network Engineering ToolOpen System InterconnectionPeripheral Component InterconnectPersonal Communications ServicePulse Code ModulationProbability Density FunctionProtocol Data UnitPiggyback RequestPhysical InterfacePassive Optical NetworkPlain Old Telephone SystemPoint to Point ProtocolPure Reservation AccessQuadrature Amplitude ModulationA-iii

Appendix AGlossary pQoSQPSKRFRFCRRRSRSARTPrtPSSASARS-CDMASCFSCFQSCTESDSLSFSL-ESPSNAPSNMPSSRCSTBSuperPONTCISTCPTelcosTDMTDMAToSTTLTSQuality of ServiceQuartenary Phase Shift KeyingRadio FrequencyRequest For CommentsReservation RequestReed SolomonRivest-Shamir-AdlemanReal Time Protocolreal-time Polling ServicesScheduling AdvanceSegmentation and ReAssembly sublayerSynchronous Code Division Multiple AccessSlot Configuration FieldSelf Clock Fair QueuingSociety of Cable Telecommunications EngineersSingle Line Digital Subscriber LineStart FieldSignalling -Link Extended Super FrameSubNetwork Attachment PointSimple Network Management ProtocolSynchronisation SourceSet Top BoxSuper Passive Optical NetworksTiming Critical Interactive ServiceTransport Control ProtocolTelephone CompaniesTime Division MultiplexingTime Division Multiple AccessType of ServiceTime To LiveTransport StreamUDPUGSUser Datagram ProtocolUnsolicited Grant SlotA-iv

Appendix AGlossary pUSBUWVDSLVoDVoIPVSBWANWWWUniversal Serial BusUnique WordVery high data rate Digital Subscriber LineVideo On DemandVoice over IPVestigial Side BandWide Area NetworkWorld Wide WebA-v

APPENDIX B: DVB/DAVIC PROTOCOL STACK ANDPACKET FORMATSB.1 Protocol stackIn accordance with [34], INA and NIU support three different solutions for the upstreamchannel: Ethernet MAC bridging, Direct IP and Point to Point (PPP) and as illustrated inFigure B.1. These three solutions allow compatible and interoperable implementationsfor transmitting IP datagrams, Ethernet MAC frames and PPP frames over ATM AAL5[49] and DVB Multiprotocol Encapsulation [33]. The Direct IP solution is mandatoryfor both INA and NIU. However, the other two solutions are optional.Ethernet MAC BridgingApplication (RTP-12)TCP(20) UDP(8)IP(20)LLC (3)SNAP (5)Ethernet MAC (18)Direct IPApplication (RTP-12)TCP(20)IP(20)UDP(8)AAL5(8)ATM (5)Upstream Slot (11)PPPApplication (RTP-12)TCP(20) UDP(8)IP(20)PPP-NLPID(1) & PID (2)RTP: Real-time Transport ProtocolUDP:User Datagram ProtocolTCP: Tranfer Control ProtocolIP: Internet ProtocolSNAP: SubNetwork Attachment PointATM: Asynchronous Transfer ModeAAL5: ATM Adaptation Layer 5PID: Protocol IdentifierNLPID: Network Layer PIDMAC: Media Access Controlote: umbers in parenthesis represent the protocol overhead per data unitFigure B.1 – Protocol stack.A-vi

Appendix BProtocol stack and packet formats pB.2 Packet formatsIn the upstream channel, a 64-byte slot format is used for the transmission of user dataor MAC messages, as shown in Figure B.2a. A Unique Word (UW) of 4-bytes providesa burst mode acquisition method for synchronisation purposes. The payload area is of53-bytes and contains a single ATM cell for user data or MAC control transmissions.Then, a Reed-Solomon (RS) parity field of 6-bytes provides 3-bytes of RS protectionover the payload area. The Guard Band of 1-byte provides spacing between adjacentpackets.In the case where minislots are used, one contention-based upstream slot is divided intothree independent minislots of 21-bytes long as illustrated in Figure B.2b. Each minislotconsists of 4-byte UW, a single byte Start Field (ST), a 16-byte payload and a singlebyte Guard Band (GB), and can only be used to carry only a shortened MAC reservationrequest message when the splitting tree algorithm (introduced in Section 3.4.3) isselected.a) Upstream Slot Formatb) Minislot Format4-bytesUW53-bytes6-bytes1-bytePayload Area (ATM Cell) RS GB5-bytesATMHeader48-bytesInformation Payload64-byte Upstream Slot21-bytes 21-bytes 21-bytes 1-byteMinislot Minislot Minislot GBBursty Data TrafficMAC Control MessageUWSF3-bytes 1-byteMssg. Conf.TypePayloadMACAddress16-bytesMAC Info.ElementGB1-byteFigure B.2 – Upstream slot and minislot frame structure of the DVB/DAVIC protocol.A-vii

Appendix BProtocol stack and packet formats pThe Protocol Data Unit (PDU) structures for the upstream direction are depicted inFigure B.3. In this channel, ATM Adaptation Layer (AAL5) is used to encapsulateMAC PDUs (Ethernet MAC Frames, IP Datagrams or point-to-point -PPP packets) inATM cells. At the physical layer, these ATM cells are further encapsulated using the(64-byte) slot format (presented in Figure B.2a).In the downstream direction two signalling methods are used: in-band (IB) and out-ofband(OOB).• IB: In the IB signalling method the downstream channel is embedded in thebroadcast channel and is oriented for the EuroModem solution. This method is usedto transmit high data rates. Up to 52 Mbps can be transmitted in one 8-MHzdownstream channel with a 256-Quadataure Amplitude Modulation (QAM). Thetransmission of data packets (MAC PDUs) is based on the DVB specification fordata broadcasting- ETSI EN 301 192 [33], in which Motion Pictures Experts Group(MPEG-2) Transport Stream (TS) frames are used to encapsulate the MAC PDUs.Upstream PDU StructureMAC PDUDataAAL5CPCSPDUDataPADT8SARPDUSAR-PDU48-bytes...SAR-PDU48-bytesATMHPayload ... HPayload5 48 5 48DVB/DAVICUpstream PHYUW4ATM Cell RS GT53 6 164-byte slot...UW4ATM Cell RS GT536 164-byte slotFigure B.3 – Upstream PDU Structure of the DVB/DAVIC protocol.A-viii

Appendix BProtocol stack and packet formats pThe transmission of MAC control messages (or simply MAC messages) is also basedon MPEG-2 frames as illustrated in Figure B.4. Up to three MAC control messagescan be encapsulated into one MPEG-2 TS frame. Most of the MAC messages are 40-bytes in length and the other 68-bytes of the MPEG-2 TS frame is comprised of a 4-byte MPEG header, a 60-byte MAC Control Information (MCI) field which containsthe signalling parameter for the control of its associated upstream channels, and 4bytes reserved for future implementations. No AAL5 layer is defined for MPEG-2TS packets cells. At the physical layer, a 16-byte RS parity field is added to eachMPEG-2 TS frames to form an RS-coded packet of 204 bytes in length.• OOB: For downstream OOB, the maximum slot transmission rate is 3.088 Mbpswith Quartenary Phase Shift Keying (QPSK) modulation and a 2-MHz channelbandwidth. A higher transmission data rate is not necessary since the OOBdownstream channel is mainly used to transmit control messages. AAL5 adaptationis used to encapsulate data information and MAC control messages in ATM cells.This signalling method is mainly oriented for the Set Top Box solution.OBB uses a Signalling Link Extended Super frame (SL-ESF) framing structure basedon ATM cells. Ten ATM cells are mapped into 24 sub-frames with additionalsignalling (MCI) and error correction information. The description of the packetformat structures for this method is beyond the scope of this research, since we focusour research for the EuroModem solution, with a downstream IB signalling mode.For a full description of the downstream (and also the upstream) packet formats thereaders are referred to [34].188-bytes MPEG2 TS Frame Structure4-bytesMPEGHeader60-bytes 40-bytes 40-bytes 40-bytesMAC Control Information (MCI)(or signaling information)MACMessageMACMessageMACMessage4-bytesRsrv.UpstreamMarker3-bytesSlotNumberMAC FlagControlMACFlagsExt.Flags2-bytes 3-bytes 26-bytes 26-bytesFigure B.4 – Downstream frame structure.A-ix

APPENDIX C: CATV PENETRATION PER COUNTRYTable C. 1 – Cable TV penetration in Europe, North America and Asia. Main sourcesfrom ECCA annual meeting 2001 [7] and European Union [30].CountryPopulation(millions)Households(millions)Homes passed by cable(millions) (%)Cable Subscribers(millions) (%)Austria 8 3 1.7 56.7 1.2 70.6Belgium 10 3.9 3.9 100 3.85 98.7Czech Rep. 10.3 3.6 1.955 54.3 0.765 39.1Denmark 5.3 1.7 1.7 100 1.375 80.9Estonia 1.5 0.576 0.246 42.7 0.167 67.9E Finland 5.181 2.28 1.4 61.4 0.902 64.4U France 58.5 23.8 8.4 35.3 3.041 36.2R Germany 82 33 26 78.8 22 84.6O Greece 10.5 3.65 0.1 2.7 0.011 11P Hungary 10.1 3.8 2.45 64.5 1.7 69.4E Ireland 3.7 1.2 0.600 50 0.470 78.3Italy 57.3 20.4 0.963 4.7 0.073 7.6Luxembourg 0.429 0.185 0.135 73 0.121 90Netherlands 15.8 6.5 6.3 96.9 6.12 97.1Norway 4.45 1.8 0.9 50 0.789 87.7Poland 38.8 13.3 6.5 48.9 5 76.9Portugal 9.96 4.16 2.18 52.4 0.925 42.4Romania 22.5 8.2 6.5 79.3 2.8 43.1Slovenia 2.35 0.67 0.45 67.2 0.243 54Spain 40 11.8 2 16.9 0.25 12.5Sweden 9 4.1 2.7 65.8 2.7 100Switzerland 7.166 3.162 2.6 82.2 2.6 100UK 59.7 25 13.14 52.6 3.43 26.1OthersUSA 276 105.7 102.5 97 65 63Canada 31 11.5 10.9 95 8.4 77China 1265 333 112.5 33.8 90 80Taiwan 23 5.67 - - 4.5 80Japan 129 47 46.6 99 9.5 20.3Shaded rows represent the countries with the highest cable penetration in Europe.A-x

APPENDIX C: D VERIFICATION TEST FOR THE EXPONENTIAL BACKOFF ALGORITHMNumber of active stations: 30, Data rate: 32 Kbps IP + 9.7 Kbps VoIP, Data taken form from simulation simulaiton time at at 59.0 to to 60.0 secondsContention slot Collision 5 Data Slot Packet ArrivalMCI\Slot 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Arrival MCI\Slot 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Arrival1 5 5 5 5 5 5 5 5 5 5 5 5 5 20 49 22 29 18 14 14 15 15 15 15 19 19 19 19 20 20 20 20 262 20 2 5 5 28 28 28 28 24 24 24 24 13 13 13 13 30 25 50 14 19 2 6 21 26 26 8 8 213 25,24,23,15 3 30 30 30 30 30 30 30 30 30 30 30 30 30 30 15 51 26 22 22 22 29 29 29 29 29 29 29 29 29 29 29 294 30 30 30 30 30 30 30 22 22 22 22 22 22 22 22 52 22 29 29 29 29 29 29 29 29 29 29 18 18 18 18 145 25 23,19 22 22 22 22 22 22 22 22 22 22 22 22 22 22 23,12,19 53 14 19 19 2 2 2 2 6 6 6 6 21 21 21 21 156 24 22 22 22 22 22 22 22 22 22 22 27 27 27 27 27 22,2 54 15 20 5 26 26 26 26 22 22 22 22 20,147 22 19 27 4 4 12 12 12 12 10 10 10 10 20 20 20 55 14 26 1 18 12 26,15 20 2 2 2 2 3 3 3 3 25 25 24 24 24 26,18 56 8 15 15 20 20 20 20 20 20 20 20 20 20 20 20 20 279 2 5 18 22 22 22 22 19 19 19 19 5 57 27 9 6 20 20 20 20 20 20 20 20 20 5 5 5 5 14 14 18,2710 23 15 26 21 6 12 12 21 58 18 14 14 26 26 26 26 26 26 26 26 26 26 26 26 26 11,1111 24 2 2 5 5 5 5 5 5 5 5 5 5 5 5 5 24,10,10 59 11 26 26 26 26 26 26 26 26 26 1 1 1 1 8 8 16,312 10 5 5 5 5 5 5 5 5 5 18 18 18 18 23 23 60 16,3 27 27 9 9 9 9 6 6 6 6 6 6 6 6 6 29,513 23 23 15 15 26 26 26 26 21 21 21 21 6 6 6 18 61 5 29,27 6 6 6 6 6 6 6 6 6 6 6 6 6 18 18 16,714 23,18 15,5 26 14 6 24 24 10 10 26,14,13 62 12 16,3 7 11 11 12,27,715 1 13 10 8 8 63 11 29 27 30,4 5 5 5 5 5 5 5 5 5 5 5 30,5,416 23 18 15 9,5 26 26 14 14 14 14 23 64 16 3 5 12 12 7 7 1617 1 1 1 1 13 13 13 13 13 13 13 13 13 13 13 11,27 65 4 7 11 11 11 11 29 29 29 29 27 27 27 27 29,21,718 27,11,5 13 13 13 13 13 13 13 13 13 13 13 10 10 8 11 66 30,29,21 27 25 16 16 16 16 3 3 3 3 3 3 3 3 3 3 3 25,19,17,1819 9 25 16 8 23 23 18 18 15 15 15 15 25,16 67 19 17 18 16 5 5 4 4 4 4 7 7 14,2620 27 11 23,14 29 18 14,29,18 68 21,14 26,7 27 27 27 27 27 27 27 27 27 27 27 27 27 27 821 9 9 9 9 25 25 25 25 25 25 25 25 25 25 25 9 69 8 27 27 27 27 27 27 27 27 25 25 25 25 19 19 19 1722 9 25 25 25 25 25 25 25 25 25 25 25 16 16 16 25 70 29 30,21,7 26,14 19 19 19 19 19 19 19 19 19 19 19 19 1923 25,14 23 30,4 16 27 27 27 27 11 11 29 29 29 29 18 18 30,4 71 19 19 19 19 19 19 17 17 17 17 18 18 16 16 8 124 5 11 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 11 72 14 26 16 17 30 1 7 20 8 8 8 29 29 20,21,125 25 4 9 9 9 9 9 9 9 23 23 23 23 23 23 73 21 13 1326 30 23 23 23 23 23 23 23 23 23 23 23 23 23 23 23 25,17 74 14 14 14 14 14 14 26 26 26 26 26 26 26 26 26 30,6,28,1227 17 23 5 5 5 5 11 11 11 11 25 25 4 4 4 4 7 75 12 28 6 26 26 26 26 26 26 26 26 26 26 26 26 26 16 1628 14 25 11 7 8,1 30 30 30 30 8 76 24 17 17 17 17 17 17 30 30 30 30 1 1 1 1 1 24,1029 17 17 17 17 77 30 10 1 1 1 1 1 1 1 1 1 1 1 1 1 1 130 8 1 14 14 25 25 25 25 11 11 7 7 7 7 78 1 1 7 7 20 20 21 21 21 13 13 13 13 12 1231 79 23,7 1 21 12 12 28 28 28 28 6 6 24 24 24 24 30 30 10 2332 8 8 8 8 1 1 80 3 29 10 10 10 2933 13 13 81 23 7 1 1 21 21 29,2034 24 29 12 28 24,29,28,12 82 1 2 20 3 3 3 3 29 2935 13 13 13 13 83 29,15 22 23 23 23 23 7 7 7 7 15,22,12,18,2236 24 24 24 24 24 24 24 24 24 24 24 24 24 24 24 24,10 84 18 12 1 1 2 2 2 2 20 20 20 20 1737 10 24 24 24 24 24 24 24 29 29 12 12 12 12 28 28 85 29 17 15 19 7 22 22 19,738 24 28 28 86 22 18 18 12 12 2239 23 29 10 10 10 10 29,23 87 29 29 17 17 15 15 15 15 19 19 19 19 7 740 3 24 24 24 24 3 88 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 26,18,1541 23 23 23 23 29 29 29 29 29 29 29 29 29 29 29 11,28,2,20 89 26 18 15 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 12,18,2042 7,2 20,11 28 29 29 29 29 29 29 29 29 29 29 29 3 3 3 3 14 90 12 21,20 6 22 22 2143 15,14 15,14 91 22 26 26 26 26 18 18 18 18 15 15 15 15 15 15 15 844 11 20,2 7 28 28 92 18 8 20 15 15 15 15 15 15 15 15 15 15 15 15 15 1545 14 15 19 19 93 21 15 12 12 12 12 12 12 12 12 12 12 12 12 12 1246 11 11 7 7 7 7 7 7 7 7 7 7 7 7 22,22,26 94 5 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 17,1447 22,2 20 26 7 7 7 7 7 7 7 7 7 7 14 14 14 14 14 95 14 17,1 12 12 12 6 6 6 6 22 22 22 22 18 18 8 8 30,148 8 14 14 14 14 14 14 14 14 14 14 14 14 14 14 14 26,18,19,29 96 30 8 20 20 20 20 20 20 20 20 20 20 20 20 20 20 9A-xi

Appendix D Verification test p97 1 9 20 20 20 20 20 20 20 20 21 21 21 21 5 5 25,27 153 12 23 23 23 23 17 17 17 17 7 7 7 7 4 4 4 1398 27 17 25 11 5 5 14 14 14 14 30 30 27,11 154 13 23 27 4 28 8 8 8 8 21 21 28,27,1299 10 16 1 1 1 1 9 9 9 9 10,16 155 16,14 12 12 16,14100 1 27 27 27 27 17 17 17 25 25 11 11 11 11 29 156 12 24 13 13 23 23 23 23 23 23 23 23 23 23 23 23 23 24,10101 29 27 10 10 16 16 16 16 18 157 16 14,10 13 23 23 23 23 23 23 23 23 23 27 27 27 27 27102 18 1 1 158 27 27 27 27 27 27 27 27 27 27 27 27 27 27 27 14,14103 4 30 24 29 29 29 29 27 27 27 30,4,27,24 159 23,10 7 14 27 27 4 4 28 28 28 28 12 12 12 12 24 24 23,15,28,1104 27 18 18 18 18 18 18 18 18 18 18 18 18 18 18 18 27 160 27,15 28 29,3,1 24 24 16 16 13 13 13 13 29105 18 18 18 18 18 18 18 4 4 4 4 30 30 30 30 24 161 23 7 7 7 7 7 7 7 7 7 7 7 7 7 7 20106 25 24 24 27 27 27 27 27 27 27 27 27 27 27 27 27 25,17 162 20,15 10,3 1 2 29 7 7 7 7 7 7 7 7 14 14 14 14 14107 24 17 27 27 27 27 27 27 27 27 27 7 163 27 14 14 14 14 14 14 14 14 14 14 14 14 14 14 14 15,9108 27,7 16 8 25 25 25 25 16,8 164 3 9 20 14 14 28 28 28 28 28 28 28 28 28 28 28109 24 24 17 17 17 17 165 14,13 19 28 28 28 28 28 28 28 28 28 28 28 23 23 23 23 13,3,19110 7 27,17 16 16 8 8 8 8 17 166 1 1 2 2 2 2 29 29 29 29 29 29 29 29 29 29,22,23,21111 167 22 21,15,14,4 23 29 29 29 29 29 29 29 29 29 29 29 29 29 27 27,24112 17 27 20 24 7 7 7 7 20,24 168 29 24 27 27 27 27 27 27 27 27 27 27 27 27 27 27 27 26,12,18,15113 13 13 169 18,13 26 21,12,6,4 27 27 27 27 27 27 3 3 3 3 9 9 20 20 6114 17 17 27 27 27 27 27 27 27 27 27 27 27 20 20 23,28,12 170 27 3 20 20 19 19 19 19 22 22 22 22 23 23 23 23 23 18,9,6,21115 28 23 12 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20 9 171 6,4 18,15,9 10 23 23 23 23 23 23 23 23 23 23 23 23 23116 9 20 20 20 20 20 24 24 13 13 13 13 24,10,30,26 172 21,14 12 23 23 23 23 29 29 29 29 29 29 29 29 29 29 29 30,26,21117 26 30 24 10 28 28 28 28 23 23 12 12 12 12 173 30,9 29 29 29 29 29 29 29 29 29 29 29 24 24118 18 14 9 9 18,14 174 21 18,14,13,5 15,6 26 26 26 26 27 27 27 27 27 27 27 27 27 27 12,11,16,14,21,2119 20 23 26 26 30 30 24 24 24 24 10 10 10 10 20,23 175 16 11,2,1 26 27 27 27 27 27 27 27 27 27 27 27 27 3 3 1120 22,3 18 18 14 14 22,3 176 30 9 6,5 18 3 3 3 3 3 3 3 3 3 3 3 3 3 27121 20 20 23 23 23 23 20 177 27 2 3 3 3 3 3 3 3 10 10 10 10 12 12 12 19,27122 3 22 2 20 178 4 19,15,11,3 1 12 12 12 12 12 12 12 12 12 12 12 12 12 12 24,11123 16 15 15,16 179 28,24,5 12 12 12 12 12 21 21 16 16 26 26 26 26 26 26 28,16,9124 3 3 3 3 22 22 2 2 2 2 20 20 20 20 20 180 21,12 6 19 16 30 30 9 9 18 18 18 18 27 27 2 2 2 2 29125 12 20,19 16 16 16 16 16 16 16 16 16 16 16 16 16 16 16 12,19,8 181 3 27,18 29,24,17,9 28 2 2 2 2 2 2 2 2 2 2 2 2 2 2 17,19126 8 16 16 16 16 16 16 16 15 15 15 15 22 182 21 13 2 2 2 2 4 4 1 1 1 1 6 6 19 19 18,30127 20 22 12 12 12 12 12 12 12 12 12 12 12 12 12 12 183 17,14,12,11 30,29 27,24 9 19,15,4 5 18,6 16 16 16 16 3 3 28 28 30,4128 19 12 12 12 12 12 12 12 12 8 8 8 8 8 8 8 26,18 184 21 21 21 21 13 13 13 13 13 13 13 13 13 13 13 23129 26 18 8 8 8 8 8 8 8 8 8 8 8 8 8 8 8 18,26,30,28 185 19,14,4 21 28,23 11 30,29 6,3 13 13 13 13 13 13 13 13 13 13 28,13,3,11,14,16130 28 30 20 20 22 22 22 22 19 19 19 19 21 186 25,18,17 27 12 16 13 9 9 9 9 5 5 5 5 25,17131 21 6 26 26 26 26 18 18 18 18 18,26 187 4 28 3 30 23 19,15,13,924,7,6 21 21 21 21 21 21 11 11 7132 18 26,25 28 28 30 30 30 30 30 30 30 30 30 30 30 30 30 25,21,11 188 25 11 14 21 27 27 27 27 12 12 16 16 27,8133 11 30 30 30 30 30 30 30 30 30 21 21 21 21 6 6 13 189 27,17,8 4 4 4 4 28 28 28 28 28 28 28 28 28 15134 26 13 25 5 21 14 6 6 18 18 14 190 13,7 9 28 28 28 28 28 28 28 28 28 28 28 28 28 3 3 29135 18 4 1 7 11 11 21,4,7 191 8 29 27 3 3 3 3 30 30 23 23 25 25 25 25 11 11 30136 26 26 26 26 26 26 26 26 26 26 26 26 26 26 26 9 192 6 30,13 11 11 14 14 21 21 9 9 9 9 9 9 9 9 1137 9 26 26 26 26 26 26 26 13 13 25 25 5 5 5 5 27 193 15 14,1 11 9 9 9 9 9 9 9 9 9 9 9 9 9 9 8 13,14,16138 27 26 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 11,13 194 18,7 9 16 24 8 8 8 29 29 29 29 27 27 27 6 6 6 6 16,28,30,12,20139 13,11 21 21 21 21 21 21 21 14 14 14 14 18 18 18 18 16 195 30,28,20 19 29,12,1 15 15 15 15 15 15 11 11 11 11 11 11 11 11 26140 21 16 18 18 4 4 4 4 4 4 4 4 4 4 4 4 4 29,28 196 15,14 17 26 11 11 11 11 11 11 11 11 11 11 11 11 11 11 24,20,10141 29 28 11 4 4 4 4 4 4 4 4 4 1 1 1 1 7 23 197 12 13 29,28 20,10 1 30 9 9 16 16 24 24 19 19142 10 23,13 7 9 9 9 9 27 27 27 27 26 26 21 21 16 16 10 198 7 18,15 14 16 24 17 17 26 26 20143 30 4 16 16 29 29 29 29 28 28 11 11 11 11 30,4 199 29 20,10 17 23 12 12 12 12 13 13 1 1 30 30 30 30 23,30,25,15144 23 13 7 15 10 10 10,23,15 200 30 28,18,13 25 7 7 7 7 14 14 16 16 24 24 24 24 3145 10 30 30 30 30 4 4 4 4 201 15,14 3 29 29 29 29 29 29 29 29 29 29 29 29 29 29 20146 25 23 23 13 13 7 7 7 7 7 7 7 7 7 7 7 25,17,23,21 202 2 29 29 29 29 29 29 29 29 17 17 17 17 23 23 5147 21 23 17 7 7 7 7 7 7 7 7 7 7 7 15 15 10 10 21 203 20,5 13 23 23 30 30 30 30 30 30 25 25 3 3 3 3 25,15,14,21,22,26,23148 7 4 8 25 25 25 25 8 204 25 23,14,10 26 30 22,21 18 2 2 2 2149 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 205 28 5 20 19,15 13 13 13 13 19150 21 21 21 21 21 21 21 21 21 21 21 21 21 21 21 206 21 23 22,14 25 25 26 26 26 30 30 18 18 22,5151 21 21 23 23 23 23 23 23 23 23 23 23 23 23 23 27,4,13 207 15 18 19 30,10 28 28 28 28 5 5 20 20 13,16152 27,13 21 23 23 23 23 23 23 23 23 23 23 23 23 23 23 23 12 208 13,5 16 20 27,26 21 21 23 23 23 23 23 23 23 23 23 23 23 23 27,26,18,17A-xii

Appendix D Verification test p209 22 17 23 23 23 23 23 23 23 23 23 23 15 15 15 15 265 27 28 28 28 28 28 28 28 28 28 29 29 29 29 24 24210 26 30,13,5 27 15 18 18 18 18 18 18 18 18 18 18 18 19 19 21 266 25,16 17 28 30 30 30 30 4 4 4 4 7 7 15 15 25,17211 6 21,18 19 19 16 16 20 20 22 22 22 22 22 22 22 22 22 267 7 27 27 27 27 7212 20 22 22 22 22 22 22 22 22 22 22 22 22 22 17 9 268 25 17 17 17 17 28 28 28 28 28 28 28 28 28 28 8213 22,9 21,18 17 26 26 26 26 27 27 15 15 15 15 15 15 6 269 1 8 28 28 28 28 28 28 28 28 28 28 28 28 7 7 7 10214 14 15,13 30,5 10 6 6 6 20 20 14,26 270 10 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7215 22 21 1 18,9 20 26 1 271 7 7 7 7 25 25 25 25 1 1 8 8 8 8216 15 13 25 2 14 14 14 14 10 10 10 10 25,9,13,2 272 7 10 10 10 10 10 10 10 10 10 10 10 10 10 10 10217 14 22 22 22 22 21 21 21 21 1 1 1 1 20 20 28,2,27 273 16 13 10 10 10 10 10 10 10 13218 18 28,27 5 20 20 26 26 26 26 26 26 26 26 26 26 26 26 26 13,11,14 274 12 28 7 7 7 7 28,12219 9 11 26 26 26 26 26 26 26 26 26 15 15 13 13 13 25 16 275 16 16 16 16 13 13 13 13220 28,27 15 16,13 25 2 2 2 2 2 2 14 14 18 18 18 18 5 29,30 276 24 16,10 12 12 12 12 28 28 28 28 24,10221 14 2 29 5 5 5 5 5 5 5 5 5 5 5 5 5 5 5 11 277222 27 30,28 13 5 5 5 5 5 5 9 9 11 11 11 11 15 278 10 16 17 24 24 24 24 17,26223 9,5 16 4 11 15 14 14 2 2 29 29 29 29 13,30,4,29 279 26 23 8 23,23,8224 29,15 14 27 27 27 27 13 13 280 20 10 10 10 10 16 16 17 17 17 17 17 17 20,3,30225 13 9,5 16 16 16 16 4 4 4 4 11 11 281 3 30 26 26 26 26 26 26 26 26 26 26 26 26 26 26 26 19,2,20226 29,25 17 15 28 30 14 14 25,17 282 19,2 26 26 26 26 26 26 26 23 23 23 23 8 8 20 20 15,9227 5 14 13 13 13 13 13 13 13 13 13 13 13 13 13 13 30 283 23,9 15 20 20 20 20 20 20 20 20 20 20 20 20 20 20 20 15,6228 7,6 26,2 13 13 13 13 13 13 13 13 17 17 17 17 15 15 26,8 284 6 19 2 20 20 20 20 20 3 3 3 3 30 30229 25 9 13,8 15 15 28 28 30 30 5 5 5 5 14 14 14 14 15 285 20,9 23 28 15 15 19,28230 7 14 6,2 29,26 15 30 5 5 286 15 6 6 6 6 6 6 19 19 2 2 2 2 22231 13 8 25 25 25 25 9 9 9 9 287 19 22,20 23 23 23 28 28 28 28 28 28 28 28 28 28 28232 7 7 7 7 14 14 15 15 15 15 15 15 15 14,22 288 9 28 28 28 28 28 28 28 28 28 28 28 15 15 15 15 26,18,12233 22,14 15 15 15 15 15 15 15 15 15 15 15 15 15 15 15 13,4 289 26,12 22 20 24,18 19 19 19 19 24234 4 26,2 6 30 30 5 5 13 13 8 8 8 8 28,12 290 6 21 9 9 21235 14 30 22 12 9 13 29,28 9 291 26 24 18 12 22 22 22 22 20 20 20 20236 26 24 4 4 6 6 6 6 6 6 6 6 6 6 6 6 24,10 292 6 6 6 6 21 21 21 21 25237 10 28 6 6 6 6 6 6 6 6 6 6 14 14 30 30 21,4 293 25 26 26 26 26 24 24 18 18 18 18 12 12 12 12 12238 29,21 4 30 30 22 22 12 12 12 12 9 9 9 9 9 9 9 25 294 5 12 12 12 12 12 12 12 12 12 12 12 12 12 12 12 14239 25 30,23 9 9 9 9 9 9 9 9 9 9 9 9 9 9 9 23,27 295 1 23,14 12 12 12 12 12 12 12 12 12 12 12 12 25 25 23,17240 27,21 13 13 13 13 26 26 24 24 24 24 10 10 10 10 7 296 17,9 2 13 5 5 5 5 13241 7 30,2 23 3 28 28 28 28 4 4 25 25 20 297 5 23 14 30 1 1 1 1 30,27242 29 13 27 20 22 21 13,22 298 27 17 2 2 13 13 13 13 13 13 13 13 13 13 13 13 11243 15 7 7 7 7 7 7 7 7 7 7 7 7 7 7 15,29,13,10 299 11 8 9 13 13 13 13 13 13 13 13 13 13 5 5 23 23 14 16,18244 10 7 7 7 7 7 7 7 7 23 23 23 23 3 3 3 300 18 16 14 14 14 30 30 30 27 27 27 27 17 17 29,7245 2 19 3 29 29 29 29 29 29 29 29 29 29 29 29 29 29 19,22,27 301 7 29 20 11 11 11 11 8 8 8 9 9 9 9 20,5,18246 30 29 29 29 29 29 29 29 29 29 29 29 29 29 29 29 22 302 5 14 18 18 16 16 16 16 14247 6 29 29 29 13 13 13 13 13 13 27 27 20 20 20 20 26,18 303 18 30 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 30,4248 29,27,18 26 13 22 22 21 21 21 21 21 21 21 21 21 21 21 21 21 26,3,18,5,10 304 4 6 7 7 7 7 7 7 7 29 29 29 29 20 20 5 5 9249 22 5,3 21 21 21 21 21 21 21 21 21 15 15 15 15 10 10 13 305 7 9 2 27 14 14 18 18 30 30 30 30 27250 29 10 10 10 10 10 10 10 10 10 10 10 10 10 10 21 306 25 4 4 4 4 6 6 6 6 6 6 6 6 6 6 6 25,17,20251 5 3 21 27,18 10 10 10 10 10 10 2 2 19 19 19 19 30 30 307 20 17 7 7 7 7 7 7 7 7 7 7 7 7 7 7 7 29,7,3252 2 10 6 6 6 6 6 6 6 6 6 6 6 6 6 6 6 308 29,3 7 7 7 7 7 7 7 9 9 2 2 27 27 27 27 2253 6 6 6 6 6 6 6 26 26 26 13 13 22 22 309 2 16,9,8,7 27 27 27 27 27 27 27 27 27 27 27 27 27 27 27 16254 26 6 13 22 22 22 22 22 22 22 22 22 22 22 22 22 22 22 14 310 29 3 27 27 27 25 25 25 25 20 20 20 20 20 20 20 21,27255 27,1 18 14 22 22 22 22 22 29 29 29 5 5 3 3 21 21 21 13 311 27,21,8 9 20 20 20 20 20 20 20 20 20 20 20 20 20 20 11,30256 22,5 21 2 2 2 2 10 10 26 26 26 26 6 6 6 6 15,15 312 30,7 16,11 20 17 17 17 17 2 2 29 29 29 29 29 29 29 29 27257 28,15 27 9 1 13 13 18 18 14 14 14 14 28,27 313 27 21 29 29 29 29 29 29 29 29 29 29 29 29 29 29 13,8258 18 22 5 11 13 11 314 30 13 29 29 29 29 29 29 29 29 29 29 3 3 9 28,13,12,9,9259 15 28 27 27 9 9 9 9 1 1 1 1 16,24 315 11 8 12 16 28 7 9 27 27 21 21 28260 27,24,16 18 18 18 18 22 22 22 22 5 5 5 5 11 11 11 29,28,16 316 27 9 24 30 30 30 30 30 30 30 30 30 30 30 30 30 30 30 24,16,10261 29 11 13 13 13 13 13 13 13 13 13 13 13 13 13 13 317 10 30 30 30 30 30 30 30 13 13 13 13 11 11 8 8 8262 27,16 24 13 13 13 13 13 13 13 13 13 13 13 13 13 13 318 13 8 8 12 12 12 12 16 16 16 16 16 16 16 16 16 5263 30 13 13 13 13 15 15 28 28 28 28 28 28 28 28 28 30,4,7 319 23 8,5 16 16 16 16 16 16 16 16 16 16 16 16 16 28 28 23,12264 4 7 15 28 28 28 28 28 28 28 28 28 28 28 28 28 28 7 320 16,12 28 28 7 7 7 7 27 27 27 27 27 27 27 27 27 10A-xiii

APPEDIX E: RESULTS FOR A 6 MBPS UPSTREAMCHAELMean Access Delay(ms)10009008007006005004003002001000E xp. Bac koff Alg.6M bps Upstream64K bps IP Traffic38 40 42 44 46 48 50 52N umber of Active Station si2-t4 i3-t5 i4-t6 i7-t9 i8-t10 i5-t6 i5-t8System Throughput0.500.480.460.440.420.40a) Exponentinal Backoff Algorithm b) Splitting Tree AlgorithmE x p. B ac k off A lg.6M bps Ups tream64K bps IP Traffic0.3838 40 42 44 46 48 50 52N um b er of Active S tation si2-t4 i3-t5 i4-t6 i7-t9 i8-t10 i5-t6 i5-t8Mean Access Delay(ms)10009008007006005004003002001000Splt. Tree Alg.6Mbps Upstream64Kbps IP Traffic56 58 59 60 61Number of Active StationsEs-5 Es -7 Es -8 E s-10 Es -12 E s14 E s-15System Throughput0.6020.6000.5980.5960.5940.5920.5900.5880.5860.584Splt. Tree A lg.6M bps Upstream64K bps IP Traffic56 58 59 60 61 62N umber of Active S tationsE s-5 E s -7 E s-8 Es-10 Es -12 Es14 E s -15Mean Access Delay(ms)500450400350300250200150100500E x p. B ac k off A lg.6M bps Up9.6 K bps V oIP170 180 190 200 205 210 213N u m b er o f Active S tatio nsi3-t5 i5-t6 i5-t8 i5-t9 i5-t10 i7-t9 i8-t10Mean Access Delay(ms)500450400350300250200150100500Splt. Tree Alg.6M bps Ups tream9.6Kbps VoIP275 280 285 290 295 300 305N umber of Active StationsEs -5 Es-6 E s-8 E s-10 Es-14System Throughput0.330.320.310.300.290.280.270.260.25E x p. B ack off A lg.6M bps Up9.6 K bps V oIP170 180 190 200 205 210 213 220N umb er of Active S tation si3-t5 i5-t6 i5-t8 i5-t9 i5-t10 i7-t9 i8-t10System Throughput0.4550.4520.4490.4460.4430.4400.4370.4340.4310.428Splt. Tree Alg.6Mbps Upstream9.6Kbps VoIP280 285 290 295 300 305Number of Active StationsEs-5 Es-6 Es-8 Es-10 Es-1410000.65959000.60908000.55700856000.50805000.454000.40753000.35702000.306510000.25600.20 0.25 0.30 0.35 0.40 0.45 0.50 0.55 0.60 0.65Normalized Offered Load0.25 0.30 0.35 0.40 0.45 0.50 0.55 0.60 0.65Normalized Offered Load0.25 0.30 0.35 0.40 0.45 0.50 0.55 0.60 0.65Normalised Offered LoadBackoff: IP IP+VoIP VoIP Tree: IP IP+VoIP VoIP Backoff: IP IP+VoIP VoIP Tree: IP IP+VoIP VoIP Backoff: IP IP+VoIP VoIP Tree: IP IP+VoIP VoIPMean Access Delay(ms)c) Performace comparison: Exponentinal Backoff vs. Splitting Tree AlgorithmSystem ThroughputFigure E. 1 – Performance optimisation and comparison of the exponential backoff algorithmand the splitting tree algorithm, for different traffic patterns and 6 Mbps Up. Channel.A-xivUtilisation(%)

Appendix EResults for a 6 Mbps upstream channel pMean Access Delay (msec)Throughput (kbits/sec)a) Effects of packet size inisochronous streams at 8 Kbps100908070605040302010040003600320028002400200016001200AB64byte-Pk128byte-Pk256byte-Pk512byte-Pk1024byte-Pk10 60 110 160 210 260 310 360 410 460 510 560Number of Cable Modems800400064byte-Pk128byte-Pk256byte-Pk512byte-Pk1024byte-Pk8Kbps streams6.17Mbps UpstreamC8Kbps streams6.17Mbps Up10 60 110 160 210 260 310 360 410 460 510Number of Cable ModemsMean Access Delay (msec)Throughput (kbits/sec)b) Effects of packet size inisochronous streams at 32 Kbps1009080706050403020100400035003000250020001500100050032Kbps streams6.17Mbps Upstream64byte-Pk128byte-Pk256byte-Pk512byte-Pk1024byte-Pk10 20 30 40 50 60 70 80 90 100 110 120 130 140 150 160Number of Cable Modems064byte-Pk128byte-Pk256byte-Pk512byte-Pk1024byte-Pk32Kbps streams6.17Mbps Upstr10 20 30 40 50 60 70 80 90 100 110 120 130 140 150Number of Cable ModemsMean Access Delay (msec)Throughput (kbits/sec)c) Effects of packet size inisochronous streams at 64 Kbps1009080706050403020100400035003000250020001500100050064Kbps streams6.17Mbps Upstream128byte-Pk256byte-Pk512byte-Pk1024byte-Pk10 15 20 25 30 35 40 45 50 55 60 65 70Number of Cable Modems064Kbps streams6.17Mbps Upstream128byte-Pk256byte-Pk512byte-Pk1024byte-Pk10 15 20 25 30 35 40 45 50 55 60 65 70Number of Cable ModemsMean Access Delay (msec)d) Channel capacity for different isochronous streams, 128-byte packe size.1009080706050403020100128-byte pkt6.17 Mbps Up8 Kbps16 Kbps32 Kbps64 Kbps10 40 70 100 130 160 190 220 250 280 310 340Number of Cable ModemsThroughput (kbits/sec)22002000180016001400120010008006004002000128-byte pkt6.17 Mbps Up8 Kbps16 Kbps32 Kbps64 Kbps10 35 60 85 110 135 160 185 210 235 260 285Number of Cable ModemsMean Access Delay (msec)e) Fixed acces vs. reservation access1009080706050403020100Reservation Slots512-Byte Pk128-Byte Pkt10 15 20 25 30 35 40 45 50 55 60 65 70 75 80 85 90 95Number of Cable ModemsFxdSlots64Kbps streams6.17Mbps UpstreamFixed & Rsrv Access128b_Fxd512b_Fxd128b_Rsv512b_RsvMean Access Delay (msec)10090807060504030201008 Kbps16 Kbps32 Kbps64 Kbps128-byte pkt6.17 Mbps Up700 850 1000 1150 1300 1450 1600 1750 1900 2050Throughput (Kbps)Channel Utilization (%)1009080706050403020100128-byte pkt6.17 Mbps Up8 Kbps16 Kbps32 Kbps64 Kbps10 35 60 85 110 135 160 185 210 235 260 285 310 335Number of Cable ModemsThroughput (kbits/sec)440040003600320028002400200016001200800400128b_Fxd512b_Fxd128b_Rsv512bb_Rsv64Kbps streams6.17Mbps UpstreamFixed & Rsrv. Access512-byte Pkt128-byte PktRsrvSlots10 15 20 25 30 35 40 45 50 55 60 65 70 75 80Number of Cable ModemsFxdSlotf) Maximum number of streams supported for different isochronous streams and packet sizesStreams 8 Kbps 16 Kbps 32 Kbps 64 Kbps 128 KbpsPk Size Rsv Fxd Rsv Fxd Rsv Fxd Rsv Fxd Rsv Fxd64 byte 130 352 67 176 32 88 ----- 44 ----- -----128 byte 230 469 116 234 56 117 26 58 ----- 29256 byte 300 >500 144 234 70 117 35 58 17 29512 byte 360 >500 194 256 95 128 48 64 24 321024 byte 325 >500 155 256 85 128 42 64 21 32Figure E. 2 – Performance analysis of isochronous streams for different packet sizes,using the exponential backoff algorithm and a 6 Mbps upstream channel.A-xv

APPENDIX F: GUIDE TO CD-ROMThere is one CDs attached at the back cover of the thesis. The structure of the filescontained is as follows: F.1 DVB simulation model DVB_Process_Models DVB_Network_Models OPNET_Compatibility Process F.2 Publications CTS_DVB_QoS_Thesis_Paper(4) IBC_DVB99_PERFORMANCE EVALUATION_DVB_DOCSIS_Thesis_Paper(7) IBC_Tv99_DOCSIS_DVB_Comparisons_Thesis_Paper(9) IBC2001&CTE2002_DVB_Optimisation_of_CRA_Thesis_Paper(11)&(2) ICT2000_DVB_DOC_IEEE_Comparisons_Thesis_Paper(10) IEEE_BT_DVB_Performance_Evaluation_Thesis_Paper(1) IEEE_BT_DVB_Performance_Ischr_Strms_Thesis_Paper(1)_Abstract IEEE_Goblecom2001_DVB_Opmimisation_IschrStrms_Thesis_Paper(12)CTE_DVB_Opmimisation_CSA_Thesis_Paper(13) F.4 Specifications Docsis DOCSIS_MAC_Model_Thesis_Ref[77] DOCSIS1_0_RF_Interface_Spec_Thesis_Ref[21] DOCSIS1_1_RF_Interface_Spec_Thesis_Ref[22] DOCSIS1_2_RF_Interface_Spec_Thesis_Ref[23] Etsi_dvb DVB_CODING_MOD_(ETS_300_429)_Thesis_Ref[31] DVB_DATA_BROADC_(EN_301_192)_Thesis_Ref[33] DVB_DAVIC1_1_(ETSI_300_800)_Thesis_Ref[32] DVB_DAVIC1_2_(ETSI_ES_200_800)_Thesis_Ref[34] DVB_DAVIC1_3_(ETSI_200_800 )_Thesis_Ref[35] DVB_S_(EN_300_421)_Thesis_Ref[29] ETSI_TIPHON_QOS_(TR_101_329)_Thesis_Ref[117] EuroDOCSIS EuroDOCSIS_Overview_Thesis_Ref[2] EuroModem EuroModem_Thesis_Ref[36]A-xvi

Appendix FGuide to CD-ROM p IEEE802_14 IEEE_802_14_DRAF3_REV3_Thesis_Ref[54] IEEE_802_14_FATDMA_SCDMA_Thesis_Ref[54] ITU_H323 ITU-T Rec. H.323_Thesis_Ref[59] PacketCable PacketCable_Spec_Thesis_Ref[80] F.5 Thesis documents F.6 Tools Acrobat-Reader-4.0 Ghost-View22The DVB Simulation model directory contains the OPNET model used for theperformance analysis and optimisations of the DVB/DAVIC protocol. The Publicationsdirectory contains the papers authored in the course of this work in PDF format. Papersthat have been submitted for publication only contain the abstract. The Specificationsdirectory contains the specifications for CATV data protocols presented in Chapters 1, 2and 3, which are available in the public domain. Other copyrighted documents such asITU-T documents are not included.The Thesis Documents directory contains: an electronic version of this document inpostscript and pdf format and the excel file used for the theoretical analysis of theresults presented in Section 4.5. Finally, the Tools directory contains Acrobat Reader5.0 and Ghostscript, for reading and printing Acrobat and postscript files underWindows 95/98TM and NT.A-xvii

DOWNLOAD MY Ph.D Thesis - UNAM

Create successful ePaper yourself

Delete template?

Save as template?