Proceedngs of the Internatonal MultConference of Engneers and Computer Scentsts 03 Vol II, IMECS 03, March 3-5, 03, Hong Kong Pollng Cycle me Analyss for Wated-based DBA n GPOs I-Shyan Hwang, Jhong-Yue Lee, zu-ju Yeh Abstract As broadband access s evolvng from dgtal subscrber lnes to optcal access networks, the Ggabt Passve Optcal etwork (GPO), can provde huge bandwdth capacty, low cost, smple archtecture and easy mantenance, have emerged as one of the most promsng access network technologes for future last-mle solutons. o prevent data collson and ensure effcent transmsson, the pont-to-multpont topology of GPOs requres a tme-dvson meda access control (MAC) protocol to allocate the shared resource of a common upstream transmsson medum. herefore, Dynamc Bandwdth Allocaton (DBA) s an open and hot topc n the GPO. However, most proposed DBA plans to gnore the mpact of the maxmum cycle tme for Qualty of Servce (QoS) ensures maxmum delay and jtter of real-tme traffc, uplnk bandwdth utlzaton, drop probablty and farness n GPO upstream transmsson. In ths paper, we propose a Wated-based Dynamc Bandwdth Allocaton called WDBA whch s predct the arrvng real tme packet based on the proporton of watng tme for multple servces over GPOs. In addton to ensurng the qualty of QoS, our work focus on the fundamental problem of tradng-off between upstream channel utlzaton and maxmum pollng cycle tme wth dfferent proporton of raffc Contaners (-COs) traffc n a GPO wth multple OUs and verfy the accuracy of the analyss wth smulatons. Overall, our numercal results ndcate that the packet delay, throughput and drop probablty performance s better when the pollng cycle tme s longer; the farness s better when the pollng cycle tme s shorter. Index erms Ggabt Passve Optcal etwork (GPO), Dynamc Bandwdth Allocaton (DBA), Qualty of Servce (QoS), Pollng Cycle me, Wated-based Dynamc Bandwdth Allocaton (WDBA). I. IRODUCIO he Ggabt Passve Optcal etworks (GPOs) [] have been wdely consdered as the best canddate for next-generaton access networks snce t represents the hgh bandwdth, ncreased flexblty, broad area coverage, hgher splttng ratos, and economcally vable sharng of the expensve optcal lnks. GPO conssts of an optcal lne termnal (OL) located at the provder central offce (CO) and connect to a number of optcal network unts (OUs) at the customer premses by a sngle spltter/od, as llustrated Manuscrpt receved December 8, 0; revsed January 9, 03. I. S. Hwang s wth the Department of Informaton Communcaton/ Department of Computer Scence and Engneerng, Yuan-Ze Unversty, Chung-L 3003, awan (e-mal: shwang@saturn.yzu.edu.tw). J. Y. Lee s wth the Department of Computer Scence and Engneerng, Yuan-Ze Unversty, Chung-L 3003, awan (e-mal: jylee@saturn.yzu.edu.tw). zu-ju Yeh s wth the Department of Informaton Communcaton/Department of Computer Scence and Engneerng, Yuan-Ze Unversty, Chung-L 3003, awan (e-mal: s96607@mal.yzu.edu.tw). ISB: 978-988-95-6-8 ISS: 078-0958 (Prnt); ISS: 078-0966 (Onlne) n Fgure. Currently, GPO supports several bt rates n both channels such as asymmetrc or symmetrc combnatons, from 55 Mb/s to.5 Gb/s. In downstream, he GPO OL connects all OUs as a pont-to-multpont (PMP) archtecture, the OL transmts encrypted user traffcs over the shared bandwdth by broadcastng through the : spltter/od on a sngle wavelength (e.g. 90 and 550 nm). In upstream, a GPO s a multpont-to-pont (MPP) network. All OUs transmt ther data to the OL on a common wavelength (e.g. 30 nm) through the : passve combner. he man problem n the lnk layer of PO networks s occurred n upstream drecton, as all users share the same wavelength, and a medum access control protocol (MAC) s necessary to avod collsons and effcently allocate uplnk access between packets from dfferent OUs []. herefore, the tme dvson multple access (DMA) [3] s be used to provde shared hgh-bt-rate bandwdth. In a DMA scheme, tme s dvded nto perodc cycles, and these cycles are dvded nto as many tme slots as the number of OUs whch shares the channel. As a result, each slot s dedcated to one OU and every cycle s organzed n such a way that one slot transports packets from one OU perodcally. Meanwhle, GPO OL supports dynamc broadband algorthm (DBA), makng the dstrbuton of avalable bandwdth to OU more flexble. hey adapt network capacty to the traffc condtons by changng the dstrbuton of the bandwdth assgned to each OU dependng on the current requrements. he downstream frame format s a fxed frame of 5 us. hs frame perodcty assures the global synchronzaton of the whole system. Besdes, the upstream frame has the same length and can contan nformaton of several OUs. Each upstream frame contans at least the Physcal Layer Overhead upstream (PLOu) feld. Besdes the payload, t may also contan the physcal layer operaton and admnstraton and Optcal etwork Unt OL Upstream GC frame PlOu DBRu Alloc-ID 3.. n n.. 3 Spltter/OD Downstream GC frame GC Header (PCBd) Downstream GC payload US BWmap n.. 3 n.. 3 n.. 3 Fg.. GPO Archtecture. n Upstream Data OU OUn- OUn Downstream Data IMECS 03
Proceedngs of the Internatonal MultConference of Engneers and Computer Scentsts 03 Vol II, IMECS 03, March 3-5, 03, Hong Kong management upstream (PLOAMu), the power levelng sequence upstream (PLSu), and the dynamc bandwdth report upstream (DBRu) sectons. Furthermore, the GPO specfcaton defnes three ways n whch one OU can nform the OL about ts status: sendng pggy-backed reports n the upstream DBRu feld, usng status ndcaton bts n the PLOu feld, or ncludng an optonal OU report n the payload. On the other hand, the downstream frame conssts of the physcal control block downstream (PCBd) feld, the AM partton, and the GPO encapsulaton method (GEM) partton. he OL sends ponters n the PCBd feld, each of them ndcatng the tme at whch each OU starts and ends ts upstream transmsson. hs performance allows that only one OU can access the shared channel at the same tme. he FSA once seek to control each dfferent traffc stream by means of the MAC protocol to be able to affect the SLA and provde the requred qualty per user and stream for the QoS support. o the end, logcally separate queung s employed for each flow n each dfferent OU down to a fne level of resoluton. he QoS class s determned by assgnng each queue, such as Alloc-ID, to one of fve traffc contaners (-COs) that follow dfferent servce polces []: -CO s based on unsolcted perodc permts grantng fxed payload allocatons. hs s the only statc -CO not servced by DBA. -CO s ntended for VBR traffc and applcatons wth both delay and throughput requrements. he avalablty of bandwdth for the servce of ths -CO s ensured n the SLA, but ths bandwdth s assgned only on request to allow for multplexng gan. -CO3 s ntended for better than best effort servces and offers servce at a guaranteed mnmum rate; any surplus bandwdth s assgned only on request and avalablty. -CO s ntended for purely best effort servces, and as such s servced only on bandwdth avalablty up to a provsoned maxmum rate. -CO5 s a combned class of two or more of the other four -COs to remove from the MAC controller specfcaton of a target -CO when grantng access. Some papers have been nvestgated the general propertes of GPO. he greatest attenton s devoted to QoS guarantee by dynamc bandwdth allocaton n upstream. evertheless, servce provders almost always utlzed dynamc bandwdth allocaton. here s not suffcent attenton pad to nvestgaton of an nfluence of ths bandwdth allocaton on varous perod cycle tme behavors. herefore, we decded to nvestgate the propertes of maxmum pollng cycle tme n case of dynamc bandwdth allocaton. Concretely, some key QoS parameters wll be observed,.e. packet delay, jtter, throughput, drop probablty and farness. Here we focus on an mpact of dfferent maxmum pollng cycle tme on QoS parameters n GPO. he allocated bandwdth wll be shared between dfference maxmum pollng cycle tme and proporton of -COs traffc. he rest of the paper s structured as follows. Secton II ntroduces the Dynamc Bandwdth Allocaton for GPO system and secton III presents the Wated-based Dynamc Bandwdth Allocaton (WDBA) wth theoretcal explanatons. Performance evaluaton and detaled analyses are presented n Secton IV. Fnal secton concludes the paper and defnes future works. II. DYAMIC BADWIDH ALLOCAIO FOR GPO SYSEM In the GPO system, there are two forms of DBA process algorthms, whch are on Status Reportng (SR) and Status Reportng (SR) operaton, the GPO standard provdes the tools to mplement DBA and leaves the actual bandwdth allocaton scheme open to dfferent mplementatons. In SR DBA, the OL contnuously montor dle frames and surmse traffc status to allocate a small amount of extra bandwdth to each OU. OUs do not provde explct queue occupancy nformaton. Instead, the OL estmates the O queue status, typcally based on the actual transmsson n the prevous cycle. For example, f the OL observes that a gven OU s not sendng dle frames, t ncreases the bandwdth allocaton to that OU, otherwse, reduces ts allocaton accordngly. herefore, SR OUs underutlze lnk capacty, snce they do not nform queue occupancy to the OL as well as traffcs n the access network are bursty. SR DBA has the advantage that t mposes no requrements on the OU, and the dsadvantage that there s no way for the OL to know how best to assgn bandwdth across several OUs that need more. In SR DBA, All Os report ther upstream data queue occupancy, to be used by the OL calculaton process. Each O may have several -COs, each wth ts own traffc class. By combnng the queue occupancy nformaton and the provsoned SLA of each -CO, the OL can optmze the upstream bandwdth allocaton of the spare bandwdth on the PO. For all of these reasons, an effcent DBA algorthm should be SR DBA for GPO system. herefore, n ths paper, the status reportng s employed, whch deals wth the bandwdth allocaton provdng more powerful advantages. he IEEE 80.3ah ask Force (EPO) developed the multpont control protocol (MPCP), whch controls the communcaton between the downstream and the upstream channels. GPO systems use a SR DBA mechansm equvalent to EPO system. herefore, the mplementaton of a dynamc bandwdth assgnment requres the use of the MPCP protocol to dstrbute the upstream avalable bandwdth to each OU. he MPCP protocol s mplemented n the MAC layer located nsde the OL and uses fve control messages. Among them, two control messages, called Report and Gate, arbtrate the communcaton between the OL and the OUs to assgn bandwdth to each of them. In ths protocol, the OL polls OUs for ther queue status and grants bandwdth usng the MPCP GAE message, whle OUs reports ther status usng the MPCP REPOR message [5] and the length of the upstream bandwdth unts s not fxed, but dynamcally calculated dependng on the appled DBA algorthm. In general, OUs reports the backlog data at the -COs to the OL n the REPOR control messages at the front of every -CO transmsson wndow frame. he calculaton of the bandwdth assgnment can start once all the REPOR messages have been receved. Once the bandwdth assgnment and schedulng s completed, GAE messages accordng to the updated assgnment tme slot are broadcasted to the -COs, whereby data transmsson accordng to the next cycle frame can proceed. Moreover, OUs receve GAE message ncludng when and whch -CO transmts traffc data to the OL. Fgure shows a schematc vew of the GPO SR DBA process wth the OL-OU ISB: 978-988-95-6-8 ISS: 078-0958 (Prnt); ISS: 078-0966 (Onlne) IMECS 03
Proceedngs of the Internatonal MultConference of Engneers and Computer Scentsts 03 Vol II, IMECS 03, March 3-5, 03, Hong Kong -CO -CO -CO Cycle tme (n) -CO -CO -CO DBA tme OU OU OU Idle Perod Data + control message (DBA+R) REPOR message Cycle tme (n+) 5us OL -CO -CO -CO OUs GAE message -CO GAE message Fg.. Idle perod ssue n a GPO SR DBA pollng scheme. communcaton resultng n an updated bandwdth assgnment. In general, because of the non-neglgble DBA executon tme and round-trp tme leadng the dle perod problem n GPO. As shown n Fgure, the dle perod s gven by Idle Perod = DBA +R, () where R s the round-trp tme from OU to OL and DBA s the processng tme of the DBA algorthm. o elaborate, the dle perod s sum of computaton tme of DBA and the round trp tme between OL and each OU (.B. OUs cannot transmt data durng the dle perod). Hence, for the DBA scheme, reducng the dle perod becomes one of the man challenges ssues to address n order to mprove bandwdth utlzaton. Moreover, status report nformaton from an OU may be outdated wll causng another problem - queue state nconsstency due to packets that contnue to arrve durng ths watng tme. In a detaled, each OU experences a watng tme between sendng the REPOR message and sendng the buffered frames. Consequently, packets that arrve durng the watng tme and transmsson tme, have to be delayed to the next transmsson cycle, potentally leadng to stupendous packet delay and delay jtter. he above problems are even worse n ext Generaton PO (G-PO) owng to ncreased upstream rates and dstances between OL and OUs. In order to decrease packet delay and mprove overall farness, predctve schemes can be used so that traffc arrval durng the watng tme and transmsson tme s taken nto consderaton. o date, n order to decrease packet delay and mprove overall farness, varous predcton-based DBA algorthms have been proposed for GPO networks. In the predcton-based DBA algorthms, the extra bandwdth wll be allocated to each OU for the next cycle (+), the current bandwdth demand n cycle should be takes nto consderaton. he assgned bandwdth for the next cycle s calculated by OU request bandwdth demand plus a predcton term based on the algorthms of each scholars, t may be the constant credt-based [6-7], lnear credt scheme [8-0], class-based[-3] etc. However, most of the above proposals are aware of the fact that delay senstve traffc should be treated n a specalzed manner wthn the OL DBA stage. evertheless, these schemes do not address the nvestgaton of an nfluence of ths bandwdth allocaton on varous perod cycle tme behavors. In ths paper we propose a Wated-based Dynamc Bandwdth Allocaton (WDBA) ISB: 978-988-95-6-8 ISS: 078-0958 (Prnt); ISS: 078-0966 (Onlne) mechansm that uses a dfferent grantng scheme for GAE messages to mprove QoS support by predct the arrvng packet based on the proporton of watng tme and hstory request bandwdth for multple servces over GPOs. Moreover, the WDBA s adapted to varable schedulng frame sze and guarantees a mnmum bandwdth for each OU n every pollng cycle and observe that the mpact between upstream channel utlzaton and maxmum pollng cycle tme wth dfferent proporton of -CO traffc n a GPO. III. WAIED-BASED DYAMIC BADWIDH ALLOCAIO MECHAISM he motvaton of ths paper s to resolve the dle perod problem and queue state nconsstency to mprove the uplnk bandwdth utlzaton, reduce the packet latency, and provde better QoS guarantee, regardless of the envronment that whether the uplnk s under dfferent traffc load and proporton of -CO traffc. Moreover, ths work focus on the fundamental problem of tradng-off between upstream channel utlzaton and maxmum pollng cycle tme n a GPO system. o acheve ths goal, the proposed WDBA scheme, combned the wated-based predcton scheme wth Lmt Bandwdth Allocaton (LBA) and Excess Bandwdth Allocaton (EBR) scheme, apples servce level agreement (SLA) schedulng polcy[7] at each OU to guarantee Qualty of Servce (QoS) ensures maxmum delay and jtter of real-tme traffc, uplnk bandwdth utlzaton, drop probablty and farness n GPO upstream transmsson. At the same tme, an nterleaved schedulng s also ntroduced, whch s our prevous work [3] to support dfferent servces and the dfferent classes of servce requre dfferental performance bounds. he nterleaved schedulng can not only resolve the dle perod problem caused by MPCP protocol schedulng polcy- grant after report, but also reduce the packet delay and ncrease the farness between heavly-loaded OUs and lghtly-loaded OUs. Moreover, ths paper focus on the relatonshp between the maxmum cycle tme and the system performance, nstead of only to the QoS as the tradtonal DBA scheme does. he WDBA mechansm s detaled as follow. A. Interleaved Schedulng of WDBA he nterleaved schedulng of WDBA s proposed to resolve the dle perod problem and mprove bandwdth utlzaton by usng b-partton groups transmsson. he nterleaved schedulng algorthm dvdes one transmsson cycle tme nto two groups and dynamcally adjusts the bandwdth between the frst group and the second group to execute nterleaved transton to resolve the tradtonal dle perod problem. he frst group and the second group wll be performed n accordance wth the number of the OU evenly. Moreover, the -CO traffc s transmtted wth guaranteed fxed bandwdth allocaton for tme-senstve applcatons, and the -CO traffc s transmtted wth predcton mechansm for the guaranteed assured bandwdth allocaton and not tme-senstve applcatons to allevate queue state nconsstency problem; recyclng the remanng bandwdth from the frst group for the second group to obtan maxmum performance. IMECS 03
Proceedngs of the Internatonal MultConference of Engneers and Computer Scentsts 03 Vol II, IMECS 03, March 3-5, 03, Hong Kong B. Dynamc Bandwdth Allocaton of WDBA he WDBA mechansm s proposed to resolve the dle perod problem, and enhance the QoS for dfferentated servces and mprove bandwdth utlzaton by usng predcton, LBA and EBR n the GPO system. he flowchart of the WDBA mechansm s llustrated n Fgure 3, after recevng whole REPOR messages from each OU, the total avalable bandwdth B av can be calculated as Max r ( Cycle g ) 6, where r s the transmsson speed of the GPO n bts per second, Max Cycle s the maxmum cycle tme, s the number of OUs, g s the guard tme and the control message length s 6bts (Byte) [] for the GPO system. Intally, the avalable bandwdth for each group and the mnmum guaranteed bandwdth (B mn = B av /) for OU are evenly dstrbuted. After calculatng mnmum guaranteed bandwdth threshold, the WDBA executes the predcton mechansm based on the proporton of watng tme, hstorcal and current traffc status nformaton, whch s expressed n equaton (). P P P 3 R R R ( H ( R ( R ( H H watng, ), ), R R cycle, H H where P represents predct ndex R )) 3, () 0 0 represents bandwdth request of each -CO of OU, and H s the average bandwdth requrements of the hstory ten cycle of each -CO of OU, where {-CO, -CO, -CO, -CO}. For the -CO and -CO traffc, the predct ndex can be update when the R H s bgger than zero, otherwse, the predct ndex s equals to request bandwdth. Durng dynamc allocaton, the allocated tmeslot wll be adapted to the requested bandwdth. o prevent the bandwdth wasted, the lmted bandwdth allocaton mechansm follow SLA and compares the mnmum guaranteed bandwdth threshold wth the predcted ndex of each OU to get the grant bandwdth ndex (G ), whch s expressed n equaton (3). G G G mn( P mn( P mn( P, B, B, B mn mn mn ) P P ) P ). (3) In the Excessve Bandwdth Reallocaton (EBR) mechansm, the excess bandwdth can be collected from lghtly-loaded OUs and redstrbuted among the heavly-loaded OUs. he sum of underutlzed bandwdth of lghtly-loaded OUs s called excessve bandwdth (B excess ), whch can be expressed as follow: B excess jl ( B mn G ), B mn G, () where L s the set of lghtly-loaded OUs and j s a lghtly-loaded OU n L. In the end, a heavly-loaded OU obtans an addtonal bandwdth based on the EBR mechansm. If the bandwdth has not yet been dstrbuted to the heavly-loaded OUs after B excess has been allocated, the remanng avalable bandwdth (B reman ) can be reserved for the next group of OUs for DBA. he B reman s expressed n equaton (5) as follows: jh B B ( G B ), B G, (5) reman excess Start WDBA Receve REPOR message Intalze Bav and calculate Bmn Calculate predct ndex P Lmted Bandwdth Allocaton Calculate grant ndex G Calculate excess bandwdth Excessve Bandwdth Reallocaton Y Update Breman Bexcess Breman > 0 mn Fnsh WDBA mn where H s the set of heavly-loaded OUs and j s a heavly-loaded OU n H. herefore, the WDBA can support QoS and enhance system performance for dfferental servces and effcently reallocates excessve bandwdth n GPO. IV. PERFORMACE EVALUAIO In ths secton, the system performance of WDBA mechansm s compared between maxmum poolng cycle tme: mllsecond,.5 mllsecond and mllsecond n terms of the throughput, end-to-end delay, drop probablty, jtter and farness for 6 OUs. he GPO smulaton model, set up by the OPE modeler network smulator, the upstream/downstream lnk capacty s.gbps, the OL-OU dstance s 0-0km, the buffer sze s 0MByte, the guard tme s.8μs and the computaton tme of DBA s 0μs. he servce polcy follows the frst-n frst-out (FIFO) prncple. he -CO traffc has the determnstc effcacy wth lmts s antcpated. For the traffc model consdered, an extensve study has shown that most network traffc can be characterzed by self-smlarty and long-range dependence (LRD) []. he packet sze generated each tme for -CO, -CO or -CO traffc s 6, 500, 500 bytes wth probablty of 60%, 0% and 0%, respectvely [5]. he traffcs wth mnmum assured bandwdth and wth addtonal non-assured bandwdth of -CO are assumed to dstrbute evenly. In order to observe the effectve of hgh prorty traffc, the proporton of traffc profle s analyzed by smulatng the sx sgnfcant Fg. 3. Flowchart of WDBA. ISB: 978-988-95-6-8 ISS: 078-0958 (Prnt); ISS: 078-0966 (Onlne) IMECS 03
Proceedngs of the Internatonal MultConference of Engneers and Computer Scentsts 03 Vol II, IMECS 03, March 3-5, 03, Hong Kong scenaros n (-CO, -CO, -CO, and -CO) wth (0%, 60%, 0%, 0%, 6), (0%, 0%, 30%, 0%, 3), (0%, 30%, 0%, 0%, 3) and (0%, 0%, 0%, 0%, ), respectvely. he smulaton parameters are summarzed n able I. A. hroughput Fgure shows the throughput comparsons of -CO, -CO and -CO of dfferent maxmum poolng cycle tme n 6 OUs wth dfferent proportons of traffc profle for dfferent traffc loads. Fgures 3 show that the proporton of traffc wll lead greater mpact to average and -CO throughput. When the proporton of hgh prorty traffc s hgher, the average throughput s better than the others proporton rato, the man reason s that the LBA n WDBA follow the SLA polcy and the -CO has hghest prorty to get the bandwdth. Moreover, the traffc flow of -CO s hgher result n hgher bandwdth throughput of -CO. Moreover, the maxmum pollng cycle tme wll affect the system throughput when the traffc loadng exceedng 50%, and result n greater mpact of hgher proporton of hgh-prorty traffc scenaro. B. Packet delay Fgure 5 shows the packet delay comparsons of average and -CO packet delay of WDBA n 6 OUs wth dfferent proportons of traffc profle for dfferent traffc loads. he packet delay d s equal to d=d poll +d grant +d queue, packet delay(d) conssts of pollng delay(d poll ), grantng delay(d grant ) and queung delay(d queue ). Smulaton results show that the average packet delay of WDBA n ms maxmum pollng cycle tme have relatvely poor performance because of they can not transmt more data n short cycle tme. In the dfferent proporton of traffc, the scenaro 6 has the worse performance and begn to ncrease when the traffc load exceedng 50%, the scenaro 3 has the best performance, the reason s that the -CO n WDBA wll be transmtted early, so that the -CO can get more resource and hgher servce level but has worse performance n other -CO traffc. C. Drop probablty, Jtter and Farness Fgure 6 compares the drop probablty, jtter and farness of the WDBA n 6 OUs wth dfferent proportons of traffc profle for dfferent traffc loads. Smulaton results show that the blockng probablty of WDBA scheme n scenaro 3 s worst and the scenaro 3 has the best performance except traffc loadng s 00%, the reason s that scenaro 3 has lower total throughput and the allocated bandwdth of -CO and -CO must wat for the remanng unused bandwdth gven from -CO. herefore, the traffc data of -CO and -CO wll be queued n the buffer especally for -CO. he delay varance s calculated as ( d d) /, where d represents the delay tme of -CO packet and s the total number of receved -CO packets. Smulaton results show that the delay varance for -CO traffc ncreases as the traffc load ncreases especally n scenaro. he -CO jtter of WDBA s ncreasng when the traffc load exceedng 50% for able I. Smulaton scenaro umber of OUs n the 6 system Upstream/downstream lnk. Gbps capacty OL-OU dstance (unform) 0 km Buffer sze 0 MB Maxmum transmsson cycle ms,.5ms, ms tme Guard tme.8μs Computaton tme of DBA 0μs 3 raffc proporton Fg.. (a)average throughput. Fg.. (b)-co throughput. Fg. 5. Average packet delay. 0% 0% 30% 0% 0% 60% 0% 0% 0% 30% 0% 0% 0% 0% 0% 0% scenaro 6, and exceedng 70% for the others. he reason s that the transmsson order of each OU s sequental and that the -CO jtter of WDBA s depends on the cycle tme when the proporton of hgh prorty traffc s hgher the jtter gettng worse. he global farness ndex f (0 f ) has been addressed [6] whch s defned as: ISB: 978-988-95-6-8 ISS: 078-0958 (Prnt); ISS: 078-0966 (Onlne) IMECS 03
Proceedngs of the Internatonal MultConference of Engneers and Computer Scentsts 03 Vol II, IMECS 03, March 3-5, 03, Hong Kong mprovng the throughput. In performance evaluaton, the proporton of -COs and the dfferent maxmum pollng cycle tme wll lead dfferent smulaton result. he packet delay, throughput and drop probablty performance s better when the pollng cycle tme s longer; the farness s better when the pollng cycle tme s shorter. f G [ ] G [ ] Fg. 6. (a) Drop probablty. Fg. 6. (b) -CO Jtter. Fg. 6. (c) -CO Jtter., (6) where s the total number of OUs and G[] s the rato between the granted bandwdth of OU and requrement of OU. Smulaton results show that the proporton of hgh prorty traffc s hgher and the shorter maxmum pollng cycle tme wll leadng good farness. he reason s that the hgh proporton of VBR traffc wll cause sgnfcant changes n the system data flows compared wth the hgh proporton of CBR traffc. REFERECES [] IU- standardzaton (008) G.98.: Ggabt-capable passve optcal networks (GPO): General characterstcs. Avalable at http://www.tu.nt/rec/rec-g.98.-00803-p/en/. [] Alca López, oemí Merayo, Juan José Martínez, Patrca Fernández, Fber to the Home hrough Passve Optcal etworks, WDM Systems and etworks Optcal etworks, 0, pp 337-37. [3] Brunnel H (986) Message delay n DMA channels wth contguous output. IEEE rans Communcatons, Vol. 3, o.7, pp.68 68. [] John D. Angelopoulos, Helen-C. Lelgou, heodore Argyrou, and Stelos Zontos, Effcent ransport of Packets wth QoS n an FSA Algned GPO, IEEE Communcatons Magazne, February 00. [5] IEEE 80.3ah [Onlne]. Avalable: http://www.eee80.org/3/efm [6] B. Hakjeon, K. Sungchang, L. Dong-Soo and P. Chang-Soo, Dynamc bandwdth allocaton method for hgh lnk utlzaton to support SR OUs n GPO, Internatonal Conference on Advanced Communcaton echnology (ICAC), Vol., Feb. 00, pp.88-889. [7] C.H. Chang, P. Kourtesss and J.M. Senor, GPO servce level agreement based dynamc bandwdth assgnment protocol, IE Electronc Letter, Vol., o. 0, Sep. 006, pp. 73-7. [8] J. Jang, M.R. Handley and J.M. Senor, Dynamc bandwdth assgnment MAC protocol for dfferentated servces over GPO, Electroncs Letters, Vol., Issue, 006, pp. 653-655. [9] J. Jang and J.M. Senor, A new effcent dynamc MAC protocol for the delvery of multple servces over GPO, Photonc etwork Communcatons, Vol. 8, o., Aug. 009, pp.7-36. [0] J.D. Angelopoulos, H.C. Lelgou,. Argyrou and S. Zontos, Effcent transport of packets wth QoS n an FSA-algned GPO, IEEE Communcatons Magazne, Feb. 00, pp. 9-98. [] B. Skubc, B. Chen, C. Jaja, J. Ahmed and L. Wosnska, Improved scheme for estmatng -CO bandwdth demand n status reportng DBA for G-PO, Communcatons and Photoncs Conference and Exhbton (ACP), ov. 009, Asa, pp. -6. [] K. Kanonaks and I. omkos, Offset-based schedulng wth flexble ntervals for evolvng GPO networks, Journal of Lghtwave echnology, Vol. 7, o. 5, Aug. 009, pp. 359-368. [3] I.S. Hwang, J.Y. Lee and A.. Lem, A b-parttoned dynamc bandwdth allocaton mechansm for dfferentated servces to support state report OUs n GPO, Journal of Computatonal Informaton Systems, Vol. 8, o., Feb. 0, pp. 675-68. [] W. Wllnger, M.S. aqqu and A. Erramll, A bblographcal gude to self-smlar traffc and performance modelng for modern hgh-speed networks, Stochastc etworks: heory and Applcatons, Royal Statstcal Socety Lecture otes Seres, Vol., Oxford Unversty Press, 996. [5] C. Fralegh, S. Moon, B. Lyles, C. Cotton, M. Khan and D. Moll and R. Rockell, Packet-level traffc measurements from the Sprnt IP backbone, IEEE etwork, Vol. 7, Issue 6, pp. 6-6, ov.-dec. 003. [6] R. Jan, A. Durres and G. Babc, hroughput farness ndex: an explanaton, http://www.cs.wustl.edu/jan/atmf/ftp/af_f. V. COCLUSIO In ths study, mportant factors that can mprove the performance of GPO are dscussed and evaluated. he WDBA mechansm executes an nterleaved transmsson process to automatcally adjust cycle tme to resolve the dle perod problem for tradtonal DBA scheme, enhancng the system performance to reduce end-to-end packet delay and ISB: 978-988-95-6-8 ISS: 078-0958 (Prnt); ISS: 078-0966 (Onlne) IMECS 03