-%-== SERVICE DISCIPLINES PERFQRMANCE FOR WWW TRAFFIC IN GPRS SYSTEM. W. Ajib and P. Godlewski

Transcription

1 43 1 SERVICE DISCIPLINES PERFQRMANCE FOR WWW TRAFFIC IN GPRS SYSTEM W Ajib and P Gdlewski cle Natinale Sup rieure des T 1 cmmunicatins, France ABSTRACT - We investigate, in this paper, the perfrmance f service disciplines in General Packet Radi Service (GPRS) system The GPRS system is a new additinal GSM service, which prvide packet switched data services ver GSM netwrk and a packet access t data netwrks Using either best effrt service r guaranteed perfrmance service, the service discipline affects cnsiderably the MAC layer perfrmance The Wrld Wide Web (WWW) service is expected t be ne f the imprtant traffic applicatins that will emply GPRS netwrk T prvide best pssible perfrmance fr best-effrt WWW traffic in packet switching wireless netwrks, several service disciplines are prpsed We describe their mechanisms, similarities and differences These plicies perfrmance is studied in terms f thrughput, delay and lss rate by simulating the WWW traffic applicatin n uplink directin in GPRS system I Intrductin PDUs int RLC data blcks Besides, it prvides an acknwledgment mechanism based n selective repeat ARQ prtcl [SI In rder t amelirate this mechanism perfrmance, an efficient mdificatin is prpsed in [3-71 The prcedures defined at MAC layer enable multiple Mbile Statins (MS) t share a cmmn transmissin medium, which may cnsists f ne r several physical channels ie, PDCHs The physical layer prvides services fr infrmatin transfer ver physical channels between the MS and the netwrk In this paper, we centralise ur studies n MAC layer prcedures and particularly n the scheduling issue PDP = IP, X25, etc - ( a GGSN + The current data netwrks d nt assure n mre the increasing demand fr mbile data cmmunicatins and the recent develpments f mbile data applicatins In rder t satisfy these new develpments and demands, the system GPRS is currently being standardized by Eurpean Telecmmunicatins Standards Institute (ETSI) as a new additinal data bearer service f GSM The GPRS technlgy is suppsed t be an imprtant step twards the Universal Mbile Telecmmunicatins System (UMTS) Accrdingly, the GPRS perfrmance analyses have t take int accunt the UMTS ev!utin The physical channels, available in a GSM cell, are dynamically shared between GPRS service and ther GSM services and the nes assciated with GPRS are called Packet Data Channels (PDCHs) The radi interface f GPRS system is defined in [6] and an verview is given in [ The higher layer is called SubNetwrk Dependent Cnvergence Prtcl (SNDCP) It maps netwrk level characteristics nt the nes f the underlying netwrk Under the SNDCP, the Lgical Link Cntrl (LLC) layer prvides a highly reliable ciphered lgical link between the tw entities The lwer layers f GPRS radi interface are Radi Link Cntrl (RLC), Medium Access Cntrl (MAC) (defined in [l]) and the physical layers The RLC layer perfrms the segmentatin and re-assembly f LLC- Mbile Statin -%-==- Radii interface Um Figure 1 The GPRS radi interface Netwrk In figure I, the layers f radi interface in GPRS system are exhibited The basic radi packet in GPRS is the RLCiMAC data blck, called simply data blck It uses a sequence f fur timeslts n a PDCH, called blck perid, t be transmitted [I] A Temprary Blck Flw (TBF) is a physical cnnectin used t supprt the transfer f a number f data blcks T each TBF, a Temprary Flw Identity (TFI) is assigned By including the TFI in each data blck, the multiplexing f blcks destined fr (r riginated frm) different MSs n the same PDCH is perfrmed New types f packet data lgical channels are defined and mapped dynamically nt a 52-multiframe [6] They include Packet Bradcast Channels (PBCCH), Packet Cmmn 3G Mbile Cmmunicatin Technlgies, Cnference Publicatin N 471, IEE 2

2 4 Cntrl Channels (such as: Randm Access (PRACH) and Access Grant (PAGCH)) and traffic channels The material in this paper is rganized as fllws The next sectin cntains a rapid descriptin f resurce allcatin in GPRS system and the MAC prcedures fr uplink and dwnlink transmissins The third sectin specifies the scheduling pint and prpses different mechanisms The prpsed algrithms depend n the medium access mdes defined in GPRS fr uplink and dwnlink traffic They can be used fr different packet switching wireless netwrks The quality f service in GPRS and criteria f perfrmance are intrduced in the frth sectin The fifth sectin exhibits sme trafic applicatins and describes the WWW mdel used in the simulatins The QS and applicatins described in the frth and fifth sectins are deduced frm the ETSl dcuments [2-91 The simulatin mdel and results are given and discussed in sectin VI, shwing the interest f ur prpsitins Finally, cnclusins are drwning 11 Resurce allcatin in GPRS system The functins f MAC layer are related t the management f shared transmissin resurces (ie, PDCHs) and cntrl blck receptins One r mre PDCHs can be assigned t a mbile statin and ne r mre MSs can use ne PDCH The number f PDCHs, assigned t a MS, is limited by the class multislt f the mbile statin The Timing Advance Identifier (TAI) bunds the number f MSs that transmits n the same PDCH Tw MAC mdes are defined: the packet idle mde where n TBF exists and the packet transfer mde where radi resurce prviding a TBF is allcated t MS The MAC prcedures, in bth idle and transfer mdes, cncem cell re-selectin, system infrmatin bradcasting and paging prcedures The maximum number f MSs, which are in packet transfer mde mmentarily, is defined by the TFI maximum value A dwnlink transfer may be initiated by paging prcedures if the MS is in STANDBY Mbility Management (MM) state Once the MS is in "READY" MM state, the netwrk sends a packet dwnlink assignment, n PAGCH, including the list f assigned PDCHs and then it transmits data blcks perating with the RLC acknwledgment mechanism The MS mnitrs the assigned PDCHs and extracts its data blcks The uplink TBF establishment, cnsidering nly ne phase access, begins when the MS transmits an access request, n PRACH, and perfrms the access persistence cntrl mechanism The parameters f this mechanism (Max-Retrans, S, and 7') are defined in [I] and utlined in [4] A new mechanism (called Access request Retransmissin Annunced Prtcl r AQRAP) is prpsed in [4] tward an ameliratin f MAC perfrmance The netwrk respnds t the access request, n PAGCH, by a packet uplink assignment defining the medium access mechanism Upn packet assignment receipt, the MS transmits the data blcks acting with the RLC acknwledgment mechanism Three medium access mdes, fr uplink transmissin, are supprted: dynamic, extended dynamic and fixed The MS, in dynamic allcatin, mnitrs the Uplink State Flag (USF) field, cntained in dwnlink blcks, in rder t recgnise its assigned uplink blck perids The extended dynamic allcatin is a simple extensin f the dynamic ne adapted t deliver large vlume data packets Within this mde, a USF value indicates the assigned blck perids n several PDCHs In the case f the fixed allcatin, a certain amunt f assigned blck perids are fixed at the establishment f the TBF Handling with this mde, a packet uplink assignment is sent t the MS, when needed, t update the amunt f assigned resurces 111 The scheduling algrithms Only the uplink WWW traffic in GPRS cntext is cnsidered, in rder t simplify the studies Hwever the prpsed scheduling algrithms can be used als fr dwnlink traffic as well as fr varied traffic applicatins This algrithm can be furthermre used fr diversified packet switching wireless netwrks The MAC layer arbitrates the access t the shared medium between a multitude f MSs and the netwrk This layer has t manage scheduling algrithms, which define hw radi resurces are shared amng the MSs Tw scheduling algrithms have t be defined The first cncerns the distributin f PDCHs between different MSs It is executed at the cnnectin establishment This algrithm is nt the subject f this paper; s, the fllwing simple algrithm is used Cnsidering that the netwrk knws the number f MSs assigned t each PDCH and a PDCHl is suppsed mre laded than PDCHz if the MSs assigned t PDCH, are mre numerus than the nes assigned t PDCH2 When a MS, whse multislt class is x, establishes a TBF, the netwrk allcate the least laded x PDCHs ~~~~~ AI A2 A1 A, As R? A6 R3 RI A7 Aa A9 Rs ~~ I A, dnates arrival mment f the TBF, R, dnates the release mment f the TBF, Figure 2 The first scheduling algrithm Figure 2 cnsiders six PDCHs and cntains a scenari explaining the first scheduling algrithm In this figure,

3 433 a line cntains the number f mbile statins t which the PDCH is assigned where the line number gives the PDCH number Every clumn invlves the number f MSs assigned t the PDCH at the event f the clumn The symbl A; signifies the arrival mment f the TBFi and R, expresses the release event f the TBF, The secnd algrithm defines the allcatin f blck perids belnging t the same PDCH amng MSs t which this PDCH is assigned It is executed within the transmissin f data blcks This scheduling issue is als a netwrk dependent chice and the scheduling algrithm depends n the access medium allcatin mde ie, dynamic, extended dynamic r fixed Referring t [IO], we wuld prefer a scheduling algrithm t be efficient, prtective, and simple Efficient An algrithm is mre efficient than anther ne if it meets the same perfrmance under a heavier lad f traffic Prtective It is nt largely affected by abnrmal traffic cnditins such as lad netwrk fluctuatins Simple An algrithm has t be cnceptually simple t allw tractable analysis and mechanically simple t allw high speed implementatins The first scheduling algrithm, prpsed in this study, is called "dynamic" The bjective f this plicy is t divide the blc perids f a PDCH equitably between all the users f this PDCH The resurce assignment is updated at each new user arrival and at least at each blc perid This mechanism perates like a set f FIFO queues (ne queue per user) served by a Rund Rbin server whse allcatin cycle is ne blc perid It is prtective and efficient; just it is nt simple t be implemented Mrever, it needs an imprtant quantity f signaling It can be used nly with dynamic allcatin mde in uplink traffic The secnd prpsed scheduling algrithm is called First Cme First Served r FCFS It avids the interference f blcks riginated frm distinct MSs n the same PDCH An arriving set f blcks will be transmitted tgether and this user mnplizes the PDCH until the end f set transmissin The ther sets f blcks (new arriving riginated frm ther users r retransmitted blcks) have t delay in a FIFO queue accrding t their arriving When the current user releases the PDCH, the next set f blcks is served If sme blcks are negatively acknwledged, a new set f blcks is frmed and then treated as a new arriving set f blcks This algrithm is cmpatible with the fixed mde f medium access It is simple t be cnceived, analyzed and implemented but intuitively, it is nt especially eficient r prtective Within the FCFS algrithm, the retransmitted blcks grws cnsiderably the transmissin delay f a TBF because it is usually a small set f blcks which has t await the transmissin f larger set f blcks riginated frm ther users The third prpsed algrithm (called FCFS with pririty) uses apprximately the same scheduling plicy FCFS giving a transmissin pririty t the set f blcks This pririty depends nly n hw many times the blck has been sent This plicy is simple and mre efficient than the first ne The frth algrithm is called "FCFS with windws" It is characterized by an imprtant parameter (bitmap length nted W) This parameter can crrespnds t the bitmap fiend included in the packet uplink assignment in the case f fixed mde allcatin The prpsed mechanism reallcates the resurces (blc perids f a PDCH) equitably amng the users f this PDCH each "allcatin cycle'' (s Wlx blc perids, where x is the MS multislt class) In the beginning f each allcatin cycle, the next blc perids (within an allcatin cycle) are reserved equitably fr the different users The new arriving users within an allcatin cycle await the beginning f next allcatin cycle t depart the transmissin The parameter W can be adapted t the applicatin characteristics and t traffic cnditins (signaling quantity and channel quality) This mechanism can be used with the three allcatin mdes It is adapted t infrequent transmissin f large TBF and intermittent transmissins It seems prtective, efficient and simple t be implemented This algrithm is studied fr different values f the parameter, W IV The WWW applicatin mdel in GPRS The GPRS enables the cst effective and efficient use f netwrk resurces fr packet data applicatins that exhibits ne r mre f the fllwing characteristics [2]: (i) intermittent, nn peridic data transmissins, (ii) frequent transmissins f small vlumes f data and (iii) infrequent transmissins f larger vlumes f data The GPRS netwrk may supprt a set f additinal services, which include: accessing infrmatin stred in data base centers n demand nly (eg, Internet's Wrld Wide Web WWW r minitel-like), messaging services (eg, ), cnversatinal services (eg, File Transfer Prtcl (FTP) service r Internet's Telnet applicatin) and tele-actin services de (eg, credit carte validatin, lttery transactins r surveillance system) In this paper, nly the accessing infrmatin stred in database centers (ie, WWW applicatin) is cnsidered This applicatin can be mdeled as the nn real time service prpsed in [9] in rder t be used as a testperating envirnment in UMTS This mdel cnsists f a sessin arrival prcess mdel A packet service sessin cntains ne r several packet calls where a packet call crrespnds the dwnlading f a WWW dcument After the dcument is dwnlading, the user is cnsuming amunt f time called reading time A packet call cnstitutes f a bursty sequence f packets The number f packets in a packet call can be gemetrically distributed randm variable with a mean set t 25 The packet size distributin used is a Paret distributin with cut-ff S, the packet size is defined as min (MaxPS, P) where MaxPS is the maximum allwed packet size, MaxPS = bytes and P is nrmal Paret distributed randm variable and its PDF

4 is: Fr (x) = I 4 8 I,5 / x)',' Assuming that the average interarrival time between packets is small enugh t transmit all the packets f ne packet call within ne TBF, the TBF length is cnsidered as the summatin f all packets length within ne packet call 434 Multislt class f MSs: 2 S = 16, T = 8, and Max-retrans = 7 Average lad per MS: 5 kbit/s Number f traffic PDCH: 6 Number f PDCH carrying PRACH: 1 Number f PDCH carrying PAGCH: I Figures 4, 5 and 6 shws the perfrmance f the frth algrithm (FCFS with windws) using different values f the bitmap length parameter, W Three values f W are chsen: 128, 64 and 32 The value f W=32 gives the best perfrmance but implementing it demands mre f signaling prcedures and resurces v A packet arrival sessin 9, Figure 3 The WWW applicatin mdel V The criteria f perfrmance The QS parameters, which are defined as a user applicatin GPRS prfile in [2], are: service precedence, reliability, delay and thrughput The service precedence (pririty) indicates the relative pririty f maintaining the service under abnrmal cnditins The reliability indicates the transmissin characteristics requested by an applicatin ie, the prbability f lss f, duplicatin f, mis-sequencing f r crruptin f data units The delay defines the maximum values fr the mean delay and the 95-percentil delay t be incurred by data transfer thrugh GPRS netwrk The thrughput indicates the maximum bit rate and the mean bit rate requested by the user Firstly, the GPRS system will supprt principally best effrt services Therefre, nly best effrt plicies are taking int accunt in these studies The three perfrmance criteria are (i) the average glbal thrughput in the cell, which is the quantity f data received crrectly in the cell, (ii) the delay requested t transmit a TBF and (iii) the packet lss rate where a packet is suppsed lst if it is nt received crrectly The thrughput and delay cnsiders nly the TBFs crrectly received A packet crrespnds t a netwrk data unit The simulatr input is the lad in the cell, which increases by increasing the MSs number in the cell I O IS Numbe,*lM5S U1 me C d Figure 4 Thrughput vs MSs number (41h algrithm) 31!!! ;!!!! b Figure 5 The delay vs the MSs number (4'" algrithm) VI Simulatin results The simulatin parameters are: Length f a LLC frame: I52 bytes LLC acknwledgment mde: Ack Max number f retransmissin f a LLC frame: 3 RLC acknwledgment mde: Ack Max number f retransmissin f a data blck: 7 Cding channel scheme used: CS-2 Max active MSs number in the cell: 32 Max MSs assigned t a PDCH: 16 Blck errr rate (BLER) = 1% -,,,,,,, IO 12 $4 I6 18 NumberIMSsvlhecall Figure 6 Lss rate vs the MSs number (41h algrithm)

5 ! I I I I 435 Figures 7, 8 and 9 display the perfrmance cmparisn f the different prpsed scheduling algrithms This cmparisn takes int accunt the efficiency f the algrithms The dynamic scheduling algrithm is the mst efficient ne Next, the FCFS with windws algrithm is mre efficient than the FCFS and FCFS with pririty At last, The FCFS with pririty mechanism is mre efficient than the FCFS Figure 7 The thrughput vs the MSs number E e '15,,,,,,,,,,,,, I, FCFS: : : : : --m I ---I FCFS with Windws, FCFS with pririty and finally FCFS The dynamic algrithm furnishes the best efficiency Hwever, the FCFS with Windws can cmprmise the simplicity and efficiency criteria Vll Cnclusin In this paper, we discuss varied scheduling algrithms and their perfrmance is analyzed by simulatin methds The definitin f applicatins used in GPRS netwrk and these applicatins mdeling are discussed Our simulatin results shw the benefits f the prpsed algrithm and cmpare the perfrmance f the different scheduling algrithm in GPRS cntext The WWW applicatin can give a valuable mdel f infrequent transmissin f large vlume f data Frm this cause, the rule f service disciplines n GPRS perfrmance using WWW applicatin is under study In this paper, the best effrt service was studied Hwever, the intrductin f guaranteed perfrmance services in GPRS netwrk requires mre cmplicated studies f the scheduling issue REFERENCE [I] ETSI, 7/1999, GPRS, MS-BSS Interface, RLC- MAC Prtcl, ver 7 GSM ETSI, 4/99, GPRS, Service Descriptin Stagel, ver 62 GSM , [3] W Ajib and P Gdlewski, 8/99, Acknwledgment Prcedures at Radi Link Cntrl Level in GPRS ACM MSWiM 99 Seattle - I,, >A Nulrber 1 MSI YI me c~ll Figure 8 The delay vs the MSs number [4] W Ajib and P Gdlewski, 2/, A prpsal f an Access persistence Prtcl ver Data Wireless Netwrks, IEEE IPCCC Phenix [5] G Brasche and B Walke, 8/97, Cncepts, Services and Prtcls f the New GSM phase 2+ (GPRS), IEEE Cm Mag p94-14 iz g ",I 1 _,,,,,,,,,,,, : : : : d, I I ' Numhr 1 MSI in Ow sd Figure 9 The lss rate vs the MSs number The amunt f signaling needed by each algrithm crrespnds t the criteria f implementatin simplicity The scheduling algrithm classificatin, accrding t their increasing simplicity, is as fllws: Dynamic, [6] ETSI, 5/99: Overall descriptin f the GPRS radi interface, ver 62 GSM 364 [7] W Ajib and P Gdlewski, 11/99, "Acknwledgment Operatins in the RLC layer f GPRS", leee MMuC'99 San Dieg [8] J Cai and DJ Gdman, 1197, General Packet Radi Service in GSM IEEE Cm Mag p [9] UMTS, 4/1998: Selectin prcedures fr the chice f radi transmissin technlgies f the UMTS, ver 32, UMTS 33 [l] Hui Zhang, 1/1995 Service Disciplines fr Guaranteed Perfrmance Service in Packet-Switching Netwrks, Prceedings f IEEE