1 Performance Characerisaion of he MCNS OCSIS 1.0 CTV Proocol wih Prioriised Firs Come Firs Served Scheduling V.Sdralia, C.Smyhe, P.Tzerefos, S.Cvekovic bsrac The Mulimedia Cable Neworks Sysems (MCNS) aa Over Cable Service Inerface Specificaions (OCSIS) is esablished as he primary cable nework daa communicaions sandard. The head-end scheduling algorihm is no defined wihin he sandard bu i is he key funcion for providing he required performance capabiliy. Compuer simulaion, using he Common Simulaion Framework (CSF) 12 version of he CableLabs OCSIS 1.0 compuer model, has been used o predic he upsream sysem hroughpu and mean access delay. prioriised firs-come-firs-served scheduling algorihm has been considered o provide a baseline reference se of performance saisics agains which oher algorihms can be compared. The simulaions, wih heoreical confirmaion, have shown ha he maximum susainable sysem hroughpu is 1965kbps for a channel capaciy of 2560kbps and a packe size of 1500 oces, whereas for packe size of oces he maximum susainable hroughpu is only 1550kbps. The mean access delay is found o vary beween 10-900ms depending on he offered load and assuming ha he offered load does no exceed he capaciy of he channel. Excess offered load causes service sarvaion according o he assigned prioriy. Keywords: MNCS, OCSIS, HFC, performance characerisaion, FCFS scheduling algorihm, CTV neworks. 1. Inroducion Cable Companies are now exploring new echnologies which can be used o suppor digial ineracive mulimedia applicaions over heir Communiy nenna Television (CTV) infrasrucures, [1]. Over he las few years much aenion has been paid o he developmen of archiecural opions for CTV neworks ha will allow he immediae suppor of broadband services as he firs sep oward enhanced communicaion services for residenial users [2]. These echnologies range from he inroducion of well esablished Inerne devices o new access mechanisms. The curren CTV sandards aciviies, [3], are: IEEE 802.14 he IEEE commiee which is defining he broadband Meropolian rea Nework (MN) Medium ccess Conrol (MC) and Physical layer proocols [4]. The IEEE 802.14 is considering adoping OCSIS 1.1 as he IEEE 802.14c and OCSIS 1.2 as he IEEE 802.14b sandards; TM Forum Residenial roadband Working Group (RWG) invesigaing he provision of synchronous Transfer Mode (TM)o and for disribuion wihin he home iself [5]; Inerne Engineering Task Force (IETF) invesigaing he use of he Inerne Proocol (IP) Over Cable aa Neworks (IPCN). This work is based upon he use of rouers o inerconnec differen logical inerne srucures [6]; Mulimedia Cable Neworks Sysems (MCNS) parners producing he aa Over Cable Service Inerface Specificaions (OCSIS) on behalf of he Norh merican Cable indusry and using cable modem echnology. OCSIS 1.2 is adoping he IEEE 802.14 dvanced Physical Layer specificaion; igial udio-visual Council (VIC) looking a he sandards for complee end-o-end ineracive mulimedia delivery sysems [7]; Sociey of Cable elecommunicaions Engineers (SCTE) an accredied merican sandards organisaion working on compaibiliy issues for cable elecommunicaions sysems. The SCTE has successfully submied he MCNS specificaion for accepance by he ITU-T [17], [18]; The igial Video roadcasing (V) projec V has adoped he VIC recommendaions wih respec o CTV and has been responsible for he developmen of he European sandard ETS 300800 [8]. The majoriy of CTV neworks are based on he branchand-ree and ree-and-bush opologies. Modern cable neworks use boh coax and fibre opic cables for ransmission and are called Hybrid Fibre/Coax (HFC) access neworks. The HFC nework archiecure has emerged as he preferred CTV archiecure because i can readily suppor bi-direcional communicaion and signalling beween he subscriber unis and he Head-end (HE) by deploying amplifiers ha work in boh direcions. The bandwidh allocaion schemes in he US, Europe and Japan are differen bu in general he 80-450MHz region is used for broadcas analogue audio and video, 450-870MHz for downsream (or downlink or forward) communicaion and 5-70MHz for he upsream (or uplink or reurn) communicaion. The downsream and upsream pahs are divided ino 1-6 or 1-8MHz chan-
2 isribuion nework IP/LLC/SNP over IEEE 802.3 PSTN Remoe access server OCS-OSSI CMTS- RFSI O/E node CMCI ackbone nework Local configuraion server Generic backbone swich CMTS- NSI Cable Modem Terminaion Sysem (Head-end) CMTS-SMI CMTS- URFSI Tx Rx Fibre O/E node O/E node CMRI (coax) Cable modem Telco reurn IP/LLC/SNP over IEEE 802.3/Eherne/Token Ring/TM (vendor produc range) CMTRI Figure 1 The MCNS OCSIS 1.0 sysem archiecure. nels ha can carry as much as 28-40Mbps and 1.2-10Mbps respecively. The subscriber uni, called a Cable Modem (CM), is required. The basic funcion of he CM is he ranspor of daa from he cable nework o he users and from he users o he cable nework. The OCSIS 1.0 specificaions have already been adoped (version 1.1 is undergoing raificaion). I is clear ha all of he CTV daa sysems manufacurers are building OCSIS complian sysems and so i is appropriae o deermine he performance of such sysems. The performance capabiliy of OCSIS complian archiecures is dependen upon he head-end scheduling algorihms and hese are ouside of he specificaions his permis vendor differeniaion. Therefore, i is imporan o esablish a se of reference performance characerisics for he mos simple of scheduling algorihms agains which more complex algorihms can be compared. prioriised firs come firs served (p- FCFS) algorihm has been adoped as his basic scheduling algorihm o ensure ha comparison beween prioriised services is possible. The hroughpu characerisics for he basic ransmission funcions of OCSISbased cable neworks are undersood, inuiively, bu undocumened and publically unavailable. The mean access delay characerisics are no known. presen he only realisic way o evaluae he performance of OCSIS complian cable neworks is hrough compuer simulaion. There are very few measuremens from operaional OCSIS-based cable daa neworks and so, because of is funcional complexiy, here is considerable uncerainy in he performance capabiliies of such neworks. CableLabs have coordinaed he developmen of an exensive discree even simulaion model (using OPNET) of he OCSIS 1.0 medium access conrol (MC) and physical layer proocols. This model is named he Common Simulaion Framework (CSF) and version 12 was used for he work described herein. The CSF 12 conains he p-fcfs as is defaul algorihm. This work can be adoped by head-end manufacurers and operaors who will be able o compare he performance of heir sysems wih hese baseline characerisics. These comparisons will describe where a paricular scheduling algorihm implemenaion needs improvemen or where i improves upon he p-fcfs scheme. In he res of his paper we presen he operaion of he OCSIS MC proocols, summarise he CSF12 discree even simulaion model and describe he resuls obained when analysing isochronous and on-off source raffic loads for a CTV archiecure which has 200 cable modems on a single upsream channel. The source packe sizes are varied so ha he upsream sysem hroughpu, upsream hroughpu per prioriy and upsream mean access delay per service prioriy can be deermined as hey vary wih global offered load. simple heoreical model is also presened o confirm he sysem hroughpu when he offered load exceeds he upsream channel capaciy. 2. OCSIS 1.0 Specificaion In he OCSIS sysem [9] he ransmission pah over he cable nework is conrolled by he Cable Modem Terminaion Sysem (CMTS) a he HE and he Cable Modem (CM) a he cusomer premises. The reference archiecure shown in Figure 1 conains hree inerface caegories: aa inerfaces which include he CMTS Nework Side Inerface (CMTS-NSI) beween he CMTS and
3 he daa nework, specified in [10] and he CM o Cusomer premises equipmen Inerface (CMCI) beween he cusomer s compuer and he CM, specified in [11]; Operaions suppor sysems and elephone reurn pah inerfaces which correspond respecively o nework elemen managemen layer inerfaces beween he nework elemens and he high level Operaions Suppor Sysems (OSSs), defined in [12], and he inerface beween he CM and he elephone reurn pah for he cases where he reurn pah is no available or provided from he cable nework, specified in [13]; RF inerfaces, defined in [10], ha includes inerfaces beween he CM and he cable nework, he CMTS and he cable nework (in boh he upsream and downsream pahs). The principle funcion of he OCSIS v1.0 proocol specificaion is he ransparen ransfer of Inerne Proocol (IP) messages across he cable sysem (CMCI o CMTS- NSI) wihou Qualiy of Service (QoS); IPv4 suppor is mandaed bu migraion o IPv6 will be considered in laer versions. However, since he proocol operaes on bandwidh reservaion from he CMs o he CMTS, QoS can be provided by exending he proocol and using a scheduling algorihm in he HE. The proocol layers shown in Figure 2 are: he nework layer (IP), he daa link layer and he physical layer. Specifically, he daa link layer is comprised of hree sublayers: Logical Link Conrol (LLC) which conforms o he IEEE 802.2 sandard, link-securiy sublayer ha suppors he basic needs of privacy, auhorisaion and auhenicaion and he MC which suppors variable-lengh Proocol aa Unis (PUs). The physical layer is comprised of he upsream/downsream ransmission convergence (U/ TC) and he physical media dependen (PM) sublayers. The main feaures of he MC proocol are: CMTS-conrolled mix of conenion and reservaion ransmission opporuniies, a sream of minislos (unis for upsream ransmission opporuniies and an ineger muliple of 6.25µs incremens) in he upsream, bandwidh efficiency hrough suppor of variable-lengh packes, exensions provided for he fuure suppor of TM or oher PUs, suppor for muliple grades of service, suppor for a wide range of daa raes. The CMTS allocaes bandwidh via MC managemen messages (MP) which describe he uses o which he upsream mini-slos mus be pu. Since he upsream channel is modelled as a sream of minislos he CMTS mus generae he ime reference for each MP as a varying message. This consiss of a fixed header followed by a variable number of Informaion Elemens (IEs), each of which defines he slos as grans for specific saions o ransmi daa, indicaing o a CM ha is reques has been received. minimum, a MP may describe a single mini-slo and conain wo IEs: one for he daa/conenion requess and a null IE o erminae he lis. On he oher hand, a maximum, i mus no aa Link layer PHY layer CMTS-NSI Inerface o/from nework equipmen CMTS Sack IP Forwarding 802.3/IX LLC Link Securiy Cable MC U/ TC Cable PM ownsream Upsream 802.3/IX LLC Link Securiy Cable MC U/ TC Cable PM CM Sack IP Transparen ridging 802.3/IX LLC 802.3/IX MC 802.3 10aseT CMCI Inerface o/from CPE Figure 2 The MCNS OCSIS 1.0 archiecural layering. describe more han 4096 mini slos (in our model of OCSIS his value is se a 2048 o ensure he sysem is suiably responsive) and mus be bounded by a limi of 240 IEs. The basic elemens of he bandwidh allocaion scheme are: Each CM has one or more Service Ienificaions (SIs), 14 bis, and a 48-bi address; Upsream bandwidh is divided ino a sream of minislos, each one of which is numbered relaive o a maser reference mainained by he CMTS; CMs may issue requess o he CMTS for upsream bandwidh any ime ha eiher a reques or a daa PU is allowed from a paricular saion. CM, which has a packe o ransmi, mus wai unil he conenion slos, as defined in he las received MP, arrive. The CM hen aemps o reques bandwidh by accessing one of he conenion minislos. If more han one CM aemps o claim any one minislo hen he slo informaion will be corruped due o he conenion. The CM only becomes aware if i has successfully requesed bandwidh when he nex MP arrives. If he reques was successful hen he CMs SI will be idenified along wih he minislo number in which i can sar ransmission and he number of minislos assigned o i, oherwise he CM mus repea is reques aemp using he nex bach of conenion minislos. regisraion each CM declares is service prioriy and so in heavily used neworks a low prioriy CM may have access o very lile bandwidh. Wihin he head-end here are wo scheduling algorihms: he firs is responsible for upsream access scheduling wihin he MP and he second muliplexes he upsream channels ono he downsream. In boh cases he scheduling algorihm aemps o provide he requesed service access and when hese exceed capaciy o prioriise according o esablished service prioriy. Herein we are ineresed in he upsream scheduling algorihm as we assume single upsream and downsream channels wih he downsream having he greaer bandwidh. In OCSIS 1.0 he QoS provision is limied o prioriisaion and frame concaenaion (opional). In OCSIS 1.1 here is an exensive QoS definiion which is derived from he Inerne Proocol QoS
4 [14]. Some work on OCSIS scheduling algorihms has been published recenly [15], and his has focussed on comparisons beween IEEE 802.14 and OCSIS in erms of upsream conenion resoluion and he ineracion wih hree upsream scheduling algorihms. For scheduling, his work found ha under increasing heavy loads each reques should be suppored by muliple service burss hereby increasing he likelihood of piggybacking and so esablishing an accepable compromise beween daa ransmission delay and reques access delay. 3. The CableLabs OCSIS Model The OCSIS MC proocol model is being developed by CableLabs [19]. The CSF12 release of he model, used for his work, includes inerfaces, raffic sources and performance merics (since he compleion of his work CSF13 has been released). I consiss of: Nework models for modelling CTV opologies; Node models for modelling he inernal srucure of he CMs and he CMTS; Process models for modelling he CM MC and he CMTS MC process. The CSF 12 is he mos exensive model of he MCNS available and i has been used o verify he funcional behaviour of he MCNS specificaion. The curren model suppors one upsream only (he muliplexing of upsreams canno be evaluaed) and, wih he excepion of he flow of conrol informaion e.g. he MP, he downsream is no modelled (end-o-end ransfer delay canno be evaluaed). The sophisicaion of he model (some 30,000 lines of code) is exravagan for he performance analysis repored herein bu more complex analyses are also possible (and will be repored elsewhere) using his same model and comparison of hese provides greaer confidence in heir validiy. CSF 12 includes several amendmens made o he CSF 11 by he auhors, including he generic inerface for he headend scheduling algorihm. This inerface enables researchers o readily include differen head-end scheduling algorihms. The CMTS periodically generaes he managemen messages SYNC, UC and MP. The SYNC provides he CM wih global iming reference, he UC informaion abou he upsream channel and he MP he composiion of a specified upsream region. Specifically, every MP deermines he bandwidh disribuion of a fixed upsream channel using a number of IEs. The head-end scheduling algorihms are no specified in he OCSIS MC proocol; hese are considered implemenaion deails and as such will be used by manufacurers o creae produc differeniaion. In his model a simple p-fcfs algorihm is implemened oher scheduling algorihms would be expeced o show marked performance superioriy and as such he p-fcfs can be considered a reference base. par from heir SI, he CMs are idenified by heir prioriy; here are eigh prioriies (0 7, wih 0 he highes. Noe ha he OCSIS specificaion defines prioriy 7 as he highes bu he CSF currenly assigns prioriy 0 as he highes. We have adoped he CSF definiion wihin his paper as i makes no difference o he inerpreaion of he resuls unil hey are mapped o he OCSIS a which poin he prioriy values mus be reversed). Each ime a new reques arrives a he CMTS, i is queued in he corresponding lis a his prioriy. When he ime comes for he nex MP message i compues a horizon of all he evens in ime order ha can be scheduled (unil he number of IEs reaches he maximum) and hen he requess are processed in prioriy order, beginning from 0 o 7, up o a specified upper limi of slos and IEs. specified number of conenion-based reques slos are, also, allocaed by he algorihm in every MP. The MC of he CM follows he OCSIS MC specificaion and i can operae wih any CMTS scheduling algorihm. CM performs ranging a he beginning of he simulaion so as o calculae is iming offse wih reference o he CMTS. lso, when a CM has daa o send i scans he MP for a reques unicas region. If his is available, i ransmis a reques and wais for he allocaion of he requesed bandwidh in a subsequen MP. Oherwise, i searches for a reques/daa mulicas, reques broadcas or a reques/daa broadcas region in ha order; if here is no IE in he MP wihin one of hese regions, he CM wais for a subsequen MP. The assumpions of he model are: The ranging process is performed by all CMs a he beginning of he simulaion, as per he OCSIS specificaion; The CMs send requess for bandwidh o he CMTS, which hen schedules hese based on he p-fcfs scheduling. Requess are graned compleely or no a all i.e. fragmenaion is no modelled (beyond he scope of his paper); Concaenaion is no suppored hence muliple packes canno be blocked ogeher in a single bandwidh reques (beyond he scope of his paper); Conenion is allowed in conenion and reserved regions; When a CM issues a reques and is waiing for he MP allocaion response, i canno use any oher conenion opporuniies. The measuremens aken by he model are: The load offered per second from each cable modem, and hus he sysem in oal. The offered load is he number of packes for which a cable modem aemps o reques upsream bandwidh; The hroughpu per second for each cable modem (and hus prioriy as each cable modem is assigned one prioriy level), and he sysem in oal. The hroughpu is he number of daa packes received
5 a he head-end; The mean access delay for each cable modem (and hus he average mean access delay per prioriy) and he average access delay for he sysem as a whole. The mean access delay is he ime beween he CM receiving he ransfer reques and he packe being received a he head-end. 3.1 The Cable Modem Source Traffic Characerisics The nework model ha has been used consiss of 200 CMs, whose prioriies change according o he simulaion scenario (all 8 prioriies are used). The limi of 200 acive CMs was chosen because i was considered ypical of he number of acive users per head-end in an operaional sysem and required reasonable simulaion periods (each daa poin required 5-20 hours of simulaion ime). The parameer seings were: he daa backoff sar was 3 and he daa back-off end was 7; he upsream daa rae was 2560000bps, he downsream daa rae was 26970350bps and he minislo size was 32 byes. The simulaion ess used o evaluae he OCSIS 1.0 nework archiecure are divided ino wo caegories: 1. Simulaions wih isochronous raffic, generaing one message per second per CM, where he packe size was varied from o 1500 oces (he maximum packe is 1518 byes) wih incremens of byes. 25 CMs were assigned o each of he 8 prioriies; 2. Simulaions wih on-off raffic in which he packe size is consan ( byes or 1500 byes) bu he mean number of generaed messages per second per CM was varied. The values assigned o he mean on ime (when he CM sends packes o he HE), he mean iner-arrival ime and he mean off ime were se o vary he offered load. gain, 25 CMs were assigned o each prioriy. 4. iscussion 4.1 Isochronous Traffic nalysis In Figure 3, he isochronous raffic hroughpu, as denoed by he Isochronous curve, does no exceed 1965kbps (poin ) for an upsream capaciy of 2560kbps. The equivalen hroughpu per prioriy is shown in Figure 4. The hroughpu per prioriy is linear wih respec o offered load unil he load reaches 1909kbps (poin ), which gives 240kbps hroughpu per prioriy (poin ) or a sysem hroughpu of 1920kbps. The maximum sysem hroughpu is only 1960kbps (from Figure 3) and so as he offered load exceeds 1909kbps he lower prioriy services (prioriy 7 a firs, followed by prioriy 6, ec.) receive decreasing hroughpu. The prioriised FCFS HE scheduling algorihm means ha a sauraion he low prioriy services have heir bandwidh allocaion reduced. The associaed mean access delay (expressed as he mean ime beween he packe arriving a a CM s MC and i reach- 2 1700 1300 900 500 C Number of cable modems = 200 Upsream channel capaciy = 2.56Mbps Number of cable modems per prioriy = 25 Shor Pks= oces Long Pks=1500 oces Isochronous 600 1 1600 2 2600 3 3600 4 4600 Offered Load (kbps) Figure 3 Sysem hroughpu vs. offered load for varying packe size. ing he HE MC) is shown in Figure 5. For offered loads below 1909kbps he mean access delay varies beween 12-780ms for prioriy 0, and 14-900ms for prioriy 7. Wih excess offered load he mean access delay becomes 2500ms for prioriies 0-5 and considerably more for prioriies 6 and 7. When he offered load exceeds he channel capaciy he ransmission period described by he MP reaches is maximum size (2048 minislos). s he number of requess for he higher prioriies increases hen he allocaions for he lower prioriies are reduced, lowes prioriy firs, unil a prioriy becomes sarved of service, a which poin he nex lowes prioriy receives a reduced service suppor. Evenually only he highes prioriy requess will have a service. The variaion in offered load was conrolled by varying he lengh of he message and he number of acive CMs (each CM aemps o send messages of he same lengh). The maximum hroughpu is 595kbps lower han he channel capaciy due o he proocol overhead (from he medium access conrol and physical layer). sauraion, he order in which he head-end allocaes ransmission requess in he MP becomes significan; once he MP is full no furher requess can be serviced and so unserviced requess are ermed pending i.e. waiing for he nex MP allocaion. When sauraion is firs reached his means ha he las reques for he lowes prioriy will be unsuccessful and as he offered load increases he number of unsuccessful requess increases for he lowes prioriy unil none can be serviced. ha poin he nex lowes prioriy sars o receive a sarved service. In he mos exreme case only some of he highes prioriy requess will receive a service i.e. if he offered load from prioriy 0 CMs exceeds 1965kbps hen some of hose CMs will also suffer service sarvaion. The maximum load for hese ess was 200*1500*8=2.4Mbps which resuls in some sarvaion of service as he maximum fully serviced susainable offered load is 1.965Mbps i.e. 77% of capaciy. sauraion he access delay is approximaely 1s i.e. he mean service rae for each CM becomes equivalen o he mean iner-arrival rae.
6 300 270 240 The poin a which prioriy 7 raffic becomes sarved. 000 00 Prioriy 0 Prioriy 1 Prioriy 2 Prioriy 3 Prioriy 4 Prioriy 5 Prioriy 6 Prioriy 7 210 180 150 120 0 90 60 30 Prioriy 0 Prioriy 1 Prioriy 2 Prioriy 3 Prioriy 4 Prioriy 5 Prioriy 6 Prioriy 7 0 160 319 480 638 799 963 1125 1278 1443 1605 1751 1909 2093 2251 2403 C 10 160 319 480 638 799 963 1125 1278 1443 1605 1751 1909 2093 2251 2403 Figure 4 Throughpu per prioriy vs. global offered load (variable lengh isochronous raffic). Figure 5 Mean access delay per prioriy vs. global offered load (variable lengh isochronous raffic). 200 180 000 160 140 120 Prioriy 0 Prioriy 1 Prioriy 2 80 Prioriy 3 60 Prioriy 4 Prioriy 5 40 Prioriy 6 Prioriy 7 20 C E 0 297 646 989 1333 1629 1932 2219 2459 2750 3047 00 0 10 297 646 989 1333 1629 1932 2219 2459 2750 3047 Prioriy 0 Prioriy 1 Prioriy 2 Prioriy 3 Prioriy 4 Prioriy 5 Prioriy 6 Prioriy 7 Figure 6 Throughpu per prioriy vs. global offered load (on-off raffic of lengh oces). n isochronous service is mos likely o be used o suppor a Voice over IP (VoIP) service. In he case where he packes are oces in lengh, a 64kbps link (unencoded voice service) would require he ransfer of 130 packes/second i.e. a hroughpu of 104kbps or an equivalen offered load of 166.4kbps. full voice service would herefore produce an offered load of 333kbps on he CTV archiecure and, assuming his was he only load, would resul in a mean access delay in he region of 10-20ms. Given hese figures i is clear ha he proprieary Commied Informaion Rae (CIR) echniques supplied by manufacurers would be a beer mechanism for he suppor of isochronous services. 4.2 On-Off Traffic nalysis (a) Packe size equal o byes: The sysem hroughpu shown in Figure 3 is denoed by he Shor Pk curve and he equivalen hroughpu per prioriy is shown in Figure 6. The maximum upsream sysem hroughpu of he nework is 1550kbps (poin in Figure 3). This is because of he small packe size, he large number of generaed messages and he absence of concaenaion. Even when a CM has more han one packe o ransmi, i can only reques bandwidh for a single packe, 5 minislos (for he byes packe and is overhead) and consequenly, he oal number of minislos requesed Figure 7 Mean access delay per prioriy vs. global offered load (on-off raffic of lengh oces). every ime from all he CMs canno be greaer han 0 (200 CMs). This leads o coninuous grans, wih no pendings, and he same level of hroughpus for mos of he prioriies, wih he excepion of prioriy seven. In Figure 6 he wo sable hroughpus are 145kbps (poin ) for prioriy 7 and 200kbps (poin ) for he oher prioriies giving a sysem hroughpu of 1545kbps (cf. 1550kbps in Figure 3). n offered load of 1469kbps (poin C) is he load a which he hroughpus per prioriy diverge i.e. 185kbps or sysem hroughpu of 1480kbps. The equivalen mean access delay per prioriy is shown in Figure 7 in which he delay a sauraion is 47s for prioriies 0-6 (poin ) and 62s for prioriy 7 (poin ). elow sauraion he delay varies from 14-480ms as he offered load increase from 297-1469kbps, and wihin his range he delay varies significanly beween each prioriy. s shown in Figure 3, his sysem does no become sauraed i.e. he maximum size of MP is no used. Insead he hroughpu is limied by he maximum number of requess submied which canno exceed one per acive CM. The resul is ha as he offered load per CM increases he delay is increased because he requess are queued wihin he CM unil hey are served by he head-end. Figure 6 shows ha even hough he sysem hroughpu is below he maximum possible here is sill
7 700 600 500 400 300 200 E C F G H I 0 624 1294 2005 2517 2937 3338 3639 3934 4194 4432 4655 4832 4976 5106 Prioriy 0 Prioriy 1 Prioriy 2 Prioriy 3 Prioriy 4 Prioriy 5 Prioriy 6 Prioriy 7 Figure 8 Throughpu per prioriy vs. global offered load (on-off raffic of lengh 1500 oces). some differeniaion according o prioriy i.e. prioriy 7 has a lower hroughpu han prioriies 0-6. This is because he MP describes a fixed ime period in he fuure and so he nex MP mus be ransmied before he period covered by he curren MP expires. Therefore, under condiions of equal numbers of requess per prioriy, he MP will have o be ransmied before all of he prioriy 7 requess (in paricular hose ha are piggy-backed on prioriy 7 ransmied PUs) have arrived and been processed i.e. hey become pending requess (hese requess are hen placed a he end of he nex MP allocaion plan and so he cycle is repeaed). This sarvaion also accouns for he higher mean access delays shown in Figure 7. However, i is clear ha for all prioriies he mean access delay a maximum hroughpu levels (ypically 50s) are unaccepable. These delays are caused by he process of allowing one ousanding reques per CM and so reasonable delays are only possible for global offered loads under 1Mbps. In Figure 7 he fine differeniaion of he mean access delay for offered loads in he 646-1500kbps is due o he ordering of he messages wihin he MP i.e. low prioriy messages are scheduled o ransmi a he end of he defined MP period. hese loads he delay caused by he MP ordering is significan whereas a higher offered loads he delay is dominaed by he effec of muliple packes waiing in he CM. n on-off source load (his characerisic is commonly used o represen bursy raffic) for shor message lenghs has been used for hese simulaions and i is clear ha operaional sysems mus avoid hese ypes of loads if a low mean access delay is required. In cases where small packes are ypical hen he sysem mus employ concaenaion which would allow muliple packes o be ransmied in a single graned reques his would significanly reduce he average mean access delay in cases where several packes require ransmission from each CM in each MP period. Concaenaion would also increase he sysem hroughpu. (b) Packe size equal o 1500 byes: The sysem hroughpu as shown in Figure 3 is denoed by he Long Pk J 0000 000 00 0 C Prioriy 6 Prioriy 7 10 624 1294 2005 2517 2937 3338 3639 3934 4194 4432 4655 4832 4976 5106 Prioriy 0 Prioriy 1 Prioriy 2 Prioriy 3 Prioriy 4 Prioriy 5 Figure 9 Mean access delay per prioriy vs. global offered load (on-off raffic of lengh 1500 oces). curve and he equivalen hroughpu per prioriy is shown in Figure 8. The maximum upsream sysem hroughpu is 1965kbps (poin ). In Figure 8 he hroughpus per prioriy are equal unil 240kbps (poin ) giving a sysem hroughpu of 1920kbps (cf. 1960kbps in Figure 3). s he offered load increases above he channel capaciy fewer prioriies are serviced unil only prioriy 0 receives service. The gaps beween poins,, C, and E, and F, G. H, I and J increase due o fewer levels of prioriy i.e. beween poins H and I only four full services are suppored (plus a fifh which has an increasingly sarved service) whereas 6 are suppored beween poins F and G. The equivalen mean access delay per prioriy is shown in Figure 9 in which he delay for serviced prioriies above sauraion is 700ms (poin ). elow sauraion he delay varies from 25-400ms as he offered load increase from 624-1920kbps, and wihin his range he delay does no vary significanly beween each prioriy. For each service prioriy he mean access delay becomes infinie once he hroughpu per prioriy has dropped o zero. For simpliciy, his process is no shown in Figure 9 (hese poins are labelled as ). The sysem hroughpu in Figure 3 shows ha he MP is evenually sauraed i.e. he number of service requess suppored by he MP exceeds he number of acive CMs. The naure of he disribuion of he sysem hroughpu beween he differen prioriies is shown in Figure 8. Once again he service is sarved from he lower prioriies as he offered load is increased from 2-5Mbps. hese high loads he sysem is sauraed by volume bu no by number of requess i.e. here are less han wo messages waiing for ransfer a each acive CM. When a prioriy is receiving full service, and he sysem is sauraed, his leads o mean access delays which are abou 1s and as such are considerably less han hose experienced by large numbers of small messages (as shown in Figure 7). Variable daa rae services require mean access delays in he order of a few hundred milliseconds and so he global offered load should be limied o 1-1.5Mbps. In
8 Table 1 comparison of he maximum hroughpus from simulaion and heory. R c (Mbps) L p (bis) H p (bis) C s M s (bis) N N m N N p S (Mbps) 2.56 800 480 32 256 200 5 403 0 1.550 2.56 12,000 3,360 32 256 200 60 33 1980 1.968 mos neworks some 80% of he volume of informaion is carried in large packe sizes and so he characerisics shown in Figures 8 and 9 should be ypical of he services supplied o mos CTV based users. The ariff srucure would have o reflec he prioriised service as under large packe condiions he prioriisaion mechanism has a significan effec on he end-user s qualiy of service. 4.3 Sysem Throughpu Validaion In Figure 3 i is shown ha he maximum hroughpu (S ) of he cable nework canno reach he upsream bps channel capaciy (R ). This is due o physical layer overhead and unused capaciy of he channel. The hrough- c pu ha a OCSIS cable sysem can provide is given by he Equaion: S = R L p H +L N p N +C bps c p p p s Where, R Upsream daa rae (bps); c L User daa packe (bis); p H Overhead per packe (bis); p λ The mean iner-arrival rae in packes per MP map ime per CM; C Number of conenion minislos per MP; s M Size of mini-slo (bis); s N Number of CMs in he nework; N Maximum number of available minislos in a max MP (2048); N Number of mini-slos per packe, hus m N = (L + H ) / M ; m p p s N Maximum number of served requess in a MP, hus N = (N - C )/ N ; max s m N p Maximum number of mini-slos for daa plus overhead in a MP. Two condiions are o be used: when he offered load is less han he capaciy of he sysem and when he offered load exceeds he channel capaciy i.e. sauraion. In sauraion he number of served requess in he MP is less han or equal o he number of acive CMs (N N): N p = (N max - C s ) Oherwise he number of requess which could be served by a MP is greaer han he number of acive CMs (N N p >N) and so, = λ N N map m = λ N (L + H ) / M map p p s >N he maximum hroughpu is given when When N λ =1 i.e. each CM has submied a reques for daa map ransmission (each CM is permied one ousanding reques only). The accuracy of he hroughpu Equaion can be proved by solving i for he wo on-off raffic source cases using he parameers shown in Table 1 (he header values are aken from [16]). The hroughpu values from Table 1 (S) for shor and long packes compare well wih he maximum and sauraion values shown in Figure 3 i.e. 1550kbps (poin ) and 1965kbps (poin ) respecively. In Table 1 i can be seen ha for he shor message lenghs (800 bis) he maximum number of requess ha could be served by a MP is N =403, which is considerably greaer han he number of acive users N=200. Conversely only N =33 requess can be served beween he 200 acive CMs when each message is 12,000 bis long. heoreical analysis of he hroughpu per prioriy is beyond he scope of his paper. The sysem hroughpu is independen of he head-end scheduling algorihm (assuming he algorihm is capable of ensuring he upsream is loaded as heavily as he number of ransmission requess requires) whereas for he hroughpu per prioriy he capabiliies of his algorihm are he deermining facor. Therefore he resuls from he hroughpu equaion should be applicable o any suiable scheduling algorihm. 5. Conclusions The Mulimedia Cable Nework Sysems (MCNS) aa Over Cable Service Inerface Specificaions (OCSIS) for he provision of bi-direcional broadband digial daa ransfer have been widely adoped by manufacurers of
9 Communiy nenna Television (CTV) daa neworking sysems. These specificaions define he Medium ccess Conrol (MC) and Physical layer inerfaces and proocols for he ransfer of daa upsream and downsream beween he subscriber s cable modem and he service supplier s daa head-end. The downsream capaciy was se as 27Mbps and he upsream channel capaciy as 2.56Mbps. The MC proocol operaes on a conenion based reservaion sysem wih he head-end confirming, or oherwise, bandwidh allocaion via a MC managemen message (MP). The cable modems are assigned a service prioriy (from 0-7) and so he service provision is deermined wih respec o his prioriy. This qualiy of service is provided by he scheduling algorihms implemened in he head-end (hese are no defined wihin he OCSIS sandards) and for he simulaions described herein a prioriised Firs Come Firs Served (p-fcfs) algorihm has been used. The simulaion resuls show ha for an upsream channel capaciy of 2560kbps wih 200 cable modems, he maximum hroughpu (wihou concaenaion) canno exceed 1965kbps. For packe sizes of 1500 oces and variable lengh packe sizes he maximum capaciy is 1965kbps even when he offered load exceeds he upsream channel capaciy. For packe sizes of oces he maximum upsream hroughpu is 1550kbps because he 200 cable modems are no capable of uilising he maximum MP size. The difference beween he maximum hroughpu and he channel capaciy is caused by he MC and Physical layer packe overheads, unused capaciy and he muliple access scheme s MP srucure. In he case of he small packes, he lower hroughpu (1550kbps) could be improved by he use of concaenaion which would enable a cable modem o reques bandwidh for more han one packe a a ime. The maximum access delay (defined as he ime beween he packe arriving a he cable modem s MC and he head-end s MC) is ypically 12-900ms for an isochronous service wih variable lengh packes and 2.5s when his offered load exceeds he sysem s hroughpu capaciy. For on-off raffic sources he delay varies beween 14-480ms for shor lengh packes ( oces) and 25-400ms for long packes (1500 oces). sauraion (when he offered load exceeds he sysem s hroughpu capaciy) he delays become 62s (shor packes) and 700ms (long packes). The shorer long packe service is because he low prioriy services have been sarved of service whereas in he shor packe case all eigh prioriies are sill obaining a full service. From a service perspecive i is clear ha an isochronous service e.g. voice over he inerne proocol (VoIP), can only be supplied using proprieary commied informaion rae capabiliies. In he case of variable rae services, bursy, he average lengh of packes has a major effec on he sysem hroughpu and mean access delay. Small packe sizes (ypically oces) resul in reduced maximum hroughpu and very large mean access delays (ens of seconds); his delay could be reduced by using concaenaion. Long packe sizes make more efficien usage of he bandwidh (cf. classical Eherne performance capabiliies) bu under sauraion he qualiy of service received by he end-user is heavily dependen upon he prioriisaion mechanism. cknowledgemens The auhors would like o hank Norel Neworks roadband ivision, ndover, US for heir suppor of his work and in paricular G.Whie, W.Sawyer and J.Ulm. References [1] zzam,., High Speed Cable Modems, McGraw- Hill, 1997, ISN 0-07-006417-2. [2] Hernandez-Valencia J. E., rchiecures for broadband residenial IP services over CTV neworks, ell Laboraories,IEEE Nework, January/ February 1997, pp. 36-43. [3] Tzerefos, P., Smyhe, C., Sergiou, I., and Cvekovic, S., Sandards for High Speed igial Communicaions over Cable TV Neworks, Proceedings of he 6h IEE Conference on Telecommunicaions, 1998, No.451, pp.224-229. [4] IEEE Projec 802.14/a raf 3 Revision 1, IEEE 802.14 Commiee, pril 1998. [5] Residenial roadband rchiecural Framework, TM Forum Technical Commiee, F-R- 0099.000, July 1998. [6] hed M. and Roeck G., IP Over Cable aa Nework (IPCN) Service, IETF raf, 1996. [7] onnelly,. and Smyhe, C., Tuorial on he igial udio-visual Council (VIC) Sandardizaion civiy, IEE ECEJ, Vol.9, No.1, pp.46 56. [8] ETS 300800 - V Ineracion Channel for Cable TV isribuion Sysems (CTV), European Telecommunicaion Sandards Insiue, igial Video roadcasing (V), 1998. [9] aa-over-cable Service Inerface Specificaions - Cable Modem Terminaion Sysem-Nework Side Inerface Specificaion, MCNS Consorium, 1996, SP-CMTS-NSI-I01-960702. [10] aa-over-cable Service Inerface Specificaions - Cable Modem o Cusomer Premise Equipmen Inerface Specificaion, MCNS Consorium, 1996, SP-CMCI-I01-960702. [11] aa-over-cable Inerface Specificaions - Radio Frequency Inerface Specificaion, 1997, MCNS Consorium, SP-RFII01-970326. [12] aa-over-cable Service Inerface Specificaions - Operaions Suppor Sysem Inerface Specificaion, 1997, MCNS Consorium SP-OSSI-I01-970403.
10 [13] aa-over-cable Service Inerface Specificaions - Cable Modem Telephony Reurn Inerface Specificaion, 1997, MCNS Consorium SP-CMTRI-I01-970804. [14] aa-over-cable Inerface Specificaions - Radio Frequency Inerface Specificaion, 1998, MCNS Consorium, SP-RFIv1.1-01-981214. [15] ar, Y-, Huang, C-Y and Yin W-M, llocaion and Scheduling lgorihms for IEEE 802.14 and MCNS in Hybrid Fiber Coaxial Neworks, IEEE Transacions on roadcasing, Vol.44, N0.4, 1998, pp.427-435. [16] Smyhe, C., P.Tzerefos, P., S.Cvekovic, S., Sergiou, I. and Sullivan, C., Comparison of he IEEE 802.14 roadband Meropolian rea Nework Proocol and he MCNS Cable Modem Specificaions, IEE Proceedings of IC 98, Sep 1998, pp.52-57. [17] ITU-T, Recommendaion J.83 Series J: Transmission of Television, Sound Programme and Oher Mulimedia Signals, ITU-T, pril 1997. [18] ITU-T, Recommendaion J.112: Transmission Sysems for Ineracive Cable Television Services, ITU-T Pre-published, March 1999. [19] Narayanaswamy S. and Lu M., MCNS/OCSIS MC Model esign Specificaions, MIL3 Inc, 1996. iographies Polychronis Tzerefos has received he.sc. degree in Mahemaics from he risoelian Universiy of Thessaloniki in 1994 and he M.Sc. in Telemaics from he Universiy of Sheffield in 1995. He is currenly a research associae and reading for a Ph.. degree in high speed CTV neworks a he eparmen of Compuer Science, The Universiy of Sheffield. His research ineress include high speed remoe access via CTV neworks, wireless and mobile neworks, QOS over IP and raffic shaping and scheduling. Mr. Tzerefos is a suden member of he IEEE. Srba Cvekovic was born in 1959, in elgrade, Yugoslavia. He compleed his.sc. and Ph.. degrees in Elecronic and Elecrical Engineering a The Ciy Universiy, London (1983) and Universiy College London (1987), respecively. He is currenly a Reader in Telemaics in he eparmen of Compuer Science, The Universiy of Sheffield. His research ineress include mulimedia/broadband communicaion sysems, saellie/mobile communicaions, SP as well as numerical modelling and measuremens of elecromagneic fields. He has auhored over 150 publicaions in inernaional journals, conference presenaions and echnical repors. r. Cvekovic is an Honorary Member of Ea-Kappa-Nu ssociaion in he US, and an ssociae Member of he Insiuion of Elecrical Engineers, London, England. Vaia Sdralia received he Sc degree in Compuer Science from he risoelian Universiy of Thessaloniki, Greece in 1996 and he MSc in Telemaics from he Universiy of Sheffield, U.K. in 1998. She is currenly reading for a Ph.. degree in compuer science a he Universiy of Sheffield, U.K.. Her research ineress include faul recovery of CTV neworks, design and analysis of cable neworks, communicaion proocols and QOS provisioning for high-speed daa neworks. Colin Smyhe has a degree in pplied Physics and was awarded a Ph for his work on spread specrum local area neworks; boh degrees were from he Universiy of urham. uring he pas 15 years he has worked in boh he academic and he indusrial world. He has been a lecurer a he Universiies of urham and Surrey, worked for hree years a Logica Ld., was he founder and managing direcor of he sysems consancy Hyperion Sysems Ld., has aced as independen consulan o many privae and public organisaions, and has also given numerous indusrially oriened raining courses on communicaions and sofware engineering. presen he is a Professor of Compuer Science, he Universiy of Sheffield. His research ineress are in he field of LNs and WNs, inerneworking and modelling of neworks, paricularly hose based on CTV archiecures.