Rate Adaptation Impact on TCP Services Over Finite Buffer Size HSDPA Networks

Transcription

1 Canadian Journal on Multimedia & irele Network Vol., No., February 00 Rate Adatation Imat on CP Servie Over Finite Buffer Size HSDPA Network Ibraheem M. Fayed, Mohen M. antawy, Heham M. El Badawy, Salwa H. El Ramly Abtrat HSDPA i an evolution of the third generation mobile network tandard. hi i done to ahieve higher bit rate over the air link. he rooed model i ued to evaluate the effet of the CP rotool and the RNS buffer ize for different rate value on the erformane of HSDPA network. he erformane will be aeed by mean of the evaluation for both of the buffer overflow robability and the RO bloking robability. Differently from the reviou ublihed work, a novell treatment of HSDPA erformane will be done by onidering the interation between Bloking and RO robabilitie. So, the overall aket dro robability will be taken under invetigation with reet to the ued buffer ize imultaneouly with the RO ontrain. In addition, the reented work will take into onideration the erformane enhanement that may be done by deloying the rate adatation tehnique. he interation between RNS buffer ize and the CP layer i alo tudied. An analytial model will be derived to evaluate the overall dro robability and the orreondene between CP retranmiion time and normalized air uage. Keyword-CP RO; robability; HSDPA erformane; finite RNS buffer ize; rate adatation. I. INRODUCION ranmiion Control Protool (CP i the mot widely ued tranort rotool. It wa deigned for wired network where aket loe and delay are mainly aued by ongetion. For robutne, different reovery mehanim are ued with CP. It reat to ongetion with an abrut redution of the throughut (multiliatively droed to half, then take long time for the window ize to reah it normal level again. Over mobile and wirele network, oor erformane of CP are being exeriened. In uh environment, CP till interret loe and delay i.e. high round tri time (R, aued by mobility and intermittent bad hannel ondition, a ongetion indiation. A high retranmiion rate i exeriened. In ae of mall file tranfer, mot of the time i wated during the low tart and the ongetion avoidane hae []. hi aue ervie degradation and wating of reoure in wirele network where hannel are reerved for eah onnetion. A queue are intrini to the High Seed Down-link Paket Ae (HSDPA ytem, a uitable aroah to model the CP R i done by uing of equivalent queuing model []. In ae of mobile ervie, the uer are mainly downloading ontent to their mobile auing extra load to the ytem in downlink. hi will at a a otential bottlenek. he model under tudy foue on the downlink erformane. Paket dro an aear at thee bottlenek oint due to Radio Network Subytem (RNS buffer overflow or when the maximum number of retranmiion i reahed. he effet of Retranmiion ime Out (RO on the droing robability had been tudied by negleting the effet of buffering over flow and rate adatation [], [3], [4], [5]. On the other hand, the effet of ongetion due to buffer over flow during handoff roe had been tudied with out taking into onideration the effet of RO [6], [7]. he effet of rate adatation on the network throughut in CDMA network had been tudied in [8]. he reented work will ombine all of the above mentioned arameter. hi will be done via the evaluation of the overall droing robability by taking into onideration the following arameter: RNS buffer ize and it overflow robability, the rate adatation mehanim, and the RO bloking robability. But in order to not loe generality, the normalized air time will be tudied with reet to the HSDPA R air time. he aer i organized a follow: etion ΙΙ, will introdue the ytem model and it aumtion. Setion ΙΙΙ, will reent the erformane metri and it evaluation riteria. Setion IV, will reent the obtained reult and will analyze it outome. Finally, the onluion and future work are reented in etion V. II. SYSEM MODEL AND ASSUMPIONS A redued mobile network and a ingle ongetion window limited CP onnetion of a long file download roe i onidered. he network model in the urrent work onit of two end hot onneted by a bottlenek link with wire and wirele finite aaity and a finite ized dro-tail queue in the downlink diretion, and a ontant-delay link in the ulink diretion, a hown in figure (. he model aount for the roagation R and the uer-ereived bandwidth limitation. CP aket travel from ender S to lient C through a finite-ized buffer and a link of aaity (rate μ i. he rate adatation i done by withing from μ to μ at a ertain time aording to the network detined erformane. Aknowledgement (ACK- go on the ulink ath, aumulating the roagation R delay. In the reented work, CP NewReno rotool will be onidered [7], that i inluded for referene a the mot widely ued CP flavor. In the following, the main feature of the ongetion ontrol algorithm in thi rotool will be hortly ummarized. CP ongetion ontrol i window-baed; that i, the amount of outtanding data in the network i diretly ontrolled by the rotool. he intantaneou ending rate at time t an be etimated a μ(t (t/r (t where (t i the ongetion window ize and R(t i the round-tri time.

2 i B Canadian Journal on Multimedia & irele Network Vol., No., February 00 Figure. Model for uer data aket in a mobile network from the ender eretive. NewReno CP i the latet verion of the laial CP []. he ongetion ontrol i omoed of two hae, robing and dereae. Probing in turn i made u of an exonential inreae and a linear hae, low tart and ongetion avoidane, reetively. hen a timeout or three duliate ACK- notifie the ender of a lo event, the robing hae end, and the window i either halved or dereaed to. Formally, + a( on ACK reetion; where a( / ( b( after the third duliate ACK; where b( /. hi algorithm i often referred to a Additive Inreae, Multiliative Dereae (AIMD [7]. BUFFERING MODEL AND PROBABILY OF BUFFER OVERFLO CALCULAIONS During wirele tranmiion time (air time "auming that the wirele ae link i the bottlenek of the ytem", wirele buffer ouany will grow u to the maximum buffer ize, then droing to ome non-zero value and growing to maximum buffer ize again, eriodially. hen a wirele link adatation our, link bandwidth at the outut of the RNS queuing buffer dro to μ on the air link for a time at whih the ender be a ware via the ignaling onnetion in the ulink diretion with thi adatation auing an abrut eak in buffer ouany. So, there will be free buffer lot to aommodate the eak and overflow ourrene. he buffer overflow will our indeendently of the CP ender tate, thu the robability of buffer overflow equal the ratio of the buffer growth eriod when a eak of given ize ould not be aommodated to buffer growth yle eriod time. In artiular, if denote the buffer growth yle eriod time and i the eriod of time at whih a uage eak of ize B would aue an overflow, the robability of the overflow P B ( B ( PB From [7], the buffer uage eak B will be onitent with the model in [6], [7]. B ( μ μ ( and will be alulated for a ingle AIMD CP flow and the following aumtion and notation will be ued. A normalized air time (η i aumed whih i equal to the atual tranmiion time divided by the HSDPA R. he maximum buffer ize Bmax i ontant and i determined by the network oerator. he link bandwidth (rate μ i at the queue outut ha two value μ and μ and the withing between them to ahieve a eifi determined erformane at a ertain time will be ontrolled by the network adminitrator. Let min and max denote the extreme of the ongetion window (t orreonding to minimum and maximum buffer ouany, reetively, when there i no link adatation in the ytem. he model i aumed to be in ongetion avoidane hae o the reviou yle eriod i ended and the retart window ize (min will be max/ [7]. B Bmax μr and max μr + B max, otherwie the buffer would emty during the dereae hae of the well-known aw-tooth attern of (t, and CP ould not utilize the full available bandwidth. hi alo imlie that the buffer i never emty, that i B(t > 0 for all t. A the buffer i never emty, the CP ender window (t grow and hrink eriodially with B(t[7]. Minimum boundary value of B Bmax μr and minimum boundary value of max μr By onidering a ingle flow, buffer ouany at any time t i the differene of the atual ongetion window of that flow and the data not in the buffer at t, B(t (t μ R. Let then *(B denote the ritial window ize that atifie the following relation. B Bmax + B μ R (3 Hene the window ize and a buffer uage eak B aue a buffer overflow. From the buffering timing diagram whih i hown in figure (, the time interval required by the oure to be aware with the rate adatation an be alulated a following: R w + μ (4 he time required to tranmit a HSDPA frame through the air interfae with new rate (μ i μ roagation + tranmiion + terminal queuing + terminal roeing where roagation, terminal queuing and terminal roeing an be negleted with reet to tranmiion. where: μ < μ & μ > μ for bet overage ondition. he buffer ouany at the intane of rate hanging B an be evaluated from the following equation. B (0.5 R (5 B w μ μ μ

3 B Canadian Journal on Multimedia & irele Network Vol., No., February 00 Figure. Buffering model timing diagram he maximum buffer ouany Bmax an be evaluated from the ummation of the buffer uage eak B and the buffer ouany at the intane of rate hanging B. Bmax μ Rw μ + ( μ μ (6 μ hen Frame Size F μ (7 data rate ( μ μ R w F 5 F B max μ μ Rw (8 μ μ R w F + 5 F + + Bmax μ μ Rw μ μ (9 he window ize at the intant of overflow ( i evaluated baed on the reented riteria in [9], [0]: + ( dw μ μ dt ( ( μ μ ( μ + μ (0 During the ongetion avoidane hae, the loe i the ame at the intane of rate hange and at the end of the eriod. Auming i the eriod of widow inreae from min to max without rate hanging. max ( he robability of buffer overflow P B i the ratio between and and an be evaluated from ( and (. RERANSMISSION IME OU PROBABILIY (PRO If P e i the Probability of error after deoding the information blok via FEC (Forward Error Corretion, P i the robability of error after oft ombining two ueive tranmiion of the ame information blok. he number of tranmiion of a aket all er uer i a Gauian variable with mean value and variane N and σ evaluated a follow: Eah file, aket all or CP egment i tranmitted over a ertain number of ranmiion ime Interval (I "for a CP egment thi number i between and 60 I deending on the modulation and oding eleted on the radio interfae" [3], [0]. he ue of heduling on a hared hannel make the error on eah I being indeendent (the ueive I are alloated to variou uer, then the number of retranmiion of eah I data i indeendent from the other I. Uing the entral limit theorem, the um of a large number of indeendent and identially ditributed (iid ymmetri random variable an be onidered a a Gauian variable. Hene, the number of tranmiion of the aket (N ha a Gauian ditribution with mean and variane N andσ reetively. All the error are uoed to be deteted. he algorithm ued in thi work ombine only the lat two tranmiion and not all the erroneou tranmiion. hi aroah an be onidered a an aroximation, eeially that for blok error rate (BLER of 0% and ombining all the tranmiion or the lat two retranmiion lead to imilar erformane reult []. Moreover, it i aumed that the fat link adatation kee the reeived Signal to Interferene Ratio (SIR of eah tranmiion heme (modulation and oding fairly ontant (equal to SIR target. Conequently, the robability to roerly deode the aket and be error free after j (j > tranmiion i: j j Pj ( ( ( ( Note that for j, P j - P e, the mean value of the number of tranmiion of the aket N i given by + N (3 he variane for the number of tranmiion of the aket σ an be derived ( + σ (4 hee arameter are in onitene with [3]. o elet the value of P, he effetive ode rate after oft ombining two blok i τ/ []. Hene, for a given BLER (SIR target, the value of P i extrated from the urve of link level imulation, a in [3], orreonding to (modulation M, oding rate τ/ by having the ame SIR given by the urve orreonding to (modulation M, oding rate τ. Conider the oeration of N number of CP onnetion over HSDPA and the wirele ytem will be onidered in uh environment, the R of eah onnetion i R R wirele + R wired (5 herea eah CP egment i tranmitted over everal I and eah I lat for a eriod of m [4] (the delay aued by Hyer Automati Reeat requet (H-ARQ deend uon the number of retranmiion and alo on the heduling diiline being ued at HSDPA RNS. In addition, the number of retranmiion and thu I i a random variable due to the time varying nature of the hannel harateriti. In [4], it ha been aumed that heduling deend uon amount of buffered data and elaed time with an objetive of 3

4 B Canadian Journal on Multimedia & irele Network Vol., No., February 00 ahieving the required ervie bit rate for a given Uer Equiment (UE. In uh ae, robability that a I i alloated to a uer i i n (i/n t, n i the number of I when no retranmiion i needed and R wirele an be written a: n NI ( i i Rwirele j (6 n here j i the tranmiion time at ertain ervie rate (μ i.e. jμ or μ. RHSDPA an be modeled by a Gauian variable. Hene, the time needed to tranmit a CP egment R i a Gauian variable. he robability of retranmiion timeout RO exreed a Prob.(R Gauian > 0 where 0 i the firt retranmiion time out, with the Gauian aumtion lead to: ( ( x E( R P R Gauian > 0 ΕΧΡ dx R (7 δ π 0 δ ( hih i a Q funtion, o the robability of retranmiion time out (P RO an be written a: 0 E( R P RO Q δ ( R (8 By ubtituting from equation (4, (5 and equation (6 in equation (8 hene P RO i written a follow. + 0 Rwired j PRO Q (9 ( + j HE OVERALL DROPPING PROBABILIY (P he overall droing robability (P for the CP aket over HSDPA may be alulated by ombining P RO and P B at the ame j for the ame oerational ondition. So P P + P P P by uing ( and (9 hene, B RO B RO + P e + 0 Rwired j 0 Rwired j P + Q + Q + P e ( P e ( j j may be alulated from equation (0 and (. + ( μ μ max (0 ( overflow robability and robability of retranmiion time out with reet to the normalized air time (η at different buffer ize for μ 7. Mb, μ 3.6 Mb, taking the following arameter into onideration: o equal to 00 me [8], Ime and the modulation & oding heme from [], [3], [5], [7]. P e 0 - and P 0-4 from [8], [], [3], [5], [6] and R wired (50, 75 and 00 me at different buffer ize ( Bmax from 6 kbyte to Mbyte. he buffer overflow robability and the RO robability for R wired 75me, μ 7. Mb and μ 3.6Mb are illutrated in figure (3 Probability of CP RO 0 - Prto (BmaxMB P_rto (Bmax64KB P_B (BmaxMB P_B (Bmax64KB Normalized Air ime Figure 3. RO Probability and Buffer overflow robability for different buffer ize. he overall droing robability for R 75me, μ 7. Mb and μ 3.6Mb i illutrated in figure (4 Overall Droing Probability P_total(Bmax6KB P_total (Bmax3KB P_total (Bmax64KB P_total (Bmax8K P_total (Bmax56B P_total (Bmax5KB P_total (BmaxMB Buffer Overflow Bloking Probability wired And may be alulated from equation (8. Rw 5 Bmax μ + μ + μ Rw + μ ( Normalized Air ime III. NUMERICAL RESULS AND ANALYSIS Solving equation of buffering model in etion II and CP retranmiion model equation in etion III (taking into onideration a normalized fator η, numerial reult will be hown in figure for the overall droing robability, buffer Figure 4. Overall Probability for different buffer ize. In figure (3, the robability of retranmiion time out i not affeted by the maximum buffer aaity variation. hi i beaue the aaity variation doe not deend on Bmax but during the inreae of normalized air timeη, RO robability 4

5 inreae inreae Canadian Journal on Multimedia & irele Network Vol., No., February 00 inreae. he buffer overflow bloking robability P B A hown in figure (6, during the inreae of the inreae a Bmax dereae during the inreae of η. normalized air time η and dereaing the maximum buffer ize A hown in figure (4, when inreaing the normalized air Bmax the overall droing robability P inreae. he time η and dereaing the maximum buffer ize Bmax the inreaing behaviour of P i the ame for different max B. But overall droing robability P inreae. he inreaing due to the rate variation redution to.8mb (μ 3.6 Mb behaviour of P i the ame manner for different buffer ize and μ.8mb, the effet of very mall buffer ize Bmax exet for (B max B 6KB. hat i beaue, the buffer aaity 6KB dereae whih wa learly hown in figure (3. A η an not uort the buffer filling rate due to the rate variation inreae above 80% the P i the ame for different max B and from 7. Mb to 3.6Mb i.e. the buffer filling rate will be 3.6 thi i due to the inreae in the number of retranmitted Mb whih aue the buffer overflow to our more fat o aket whih imly that the RO robability i the dominant by inreaing Bmax thi overflow dereae. robability a hown in figure (5. he buffer overflow robability and the RO robability he buffer overflow robability and the RO robability for Rwired 75me, μ 3.6 Mb and μ.8mb i for Rwired 75me, μ.8 Mb and μ 0.9Mb are illutrated in figure (5. illutrated in figure (7. Probability of CP RO 0-4 Prto (BmaxMB P_rto (Bmax64KB P_B (BmaxMB 0 - P_B (Bmax64KB Normalized Air ime Buffer Overflow Bloking Parobability Probability of CP RO Prto (BmaxMB P_rto (Bmax64KB P_B (BmaxMB P_B (Bmax64KB Normalized Air ime Buffer Overflow Bloking Probability Figure 5. RO Probability and Buffer overflow robability for different buffer ize. Figure 7. RO Probability and Buffer overflow robability for different buffer ize. he overall droing robability for R wired 75me, μ 3.6 Mb and μ.8mb i illutrated in figure (6. Overall Droing Probability P_total(Bmax6KB P_total (Bmax3KB P_total (Bmax64KB P_total (Bmax8K P_total (Bmax56B P_total (Bmax5KB P_total (BmaxMB he overall droing robability for R 75me, μ.8 Mb and μ 0.9Mb i illutrated in figure (8. Overall Droing robability 0-4 P_total(Bmax6KB P_total (Bmax3KB P_total (Bmax64KB P_total (Bmax8K P_total (Bmax56B P_total (Bmax5KB P_total (BmaxMB wired Normalized Air ime Figure 6. Overall Probability for different buffer ize. In figure (5, during the inreae of normalized air time η, RO robability inreae and the inreae of Bmax ha no effet a mentioned before. he buffer overflow bloking robability P B during the inreae of normalized air time η and a Bmax dereae Normalized Air ime Figure 8. Overall Probability for different buffer ize. In figure (7, during the inreae of normalized air time η, RO robability inreae and the inreae of Bmax ha no effet a mention before. he buffer overflow bloking robability P B during the inreae of normalized air time η and a B max dereae. 5

6 Canadian Journal on Multimedia & irele Network Vol., No., February 00 A hown in figure (8, during the inreae of the normalized air time η and the maximum buffer ize Bmax the overall droing robability P dereae. he dereaing behaviour of P i the ame for different max B. hi due to the rate variation dereae to 0.9Mb (μ.8 Mb and μ 0.9Mb. A η inreae above 90% the P i the ame for different Bmax and thi i due to the inreae in the number of retranmitted aket whih imle that the RO robability be the dominant robability a hown in figure (7. he overall droing robability with different buffer ize and the orreonding normalized air time i hown in figure (9. Overall Droing Probability Figure (9 how that, in order to get lower value of P, it i reommended to kee the normalized air time below 50% for all value of ued buffer ize. In order to get lower value of P, it i reommended to kee the normalized air time below 50% for all value of ued buffer ize. raffi with different ategorie and mobility imat will be onidered in the future work. REFERENCES [] Mohamad Aaad, and Djamal Zeghlahe, CP rformane over UMS-HSDPA Sytem, t ed. Auerbah Publiation Boa Raton New York, 007. [] Levente Bodrog, Gábor Horváth, and Caba Vulkán, " Analytial CP hroughut Model for HSDPA" Nokia Siemen Network, Budaet, Hungary, aba.vulkan@nn.om, 006. [3] Mohamad Aaad, and Djamal Zeghlahe, "How to Minimize the CP Effet in a UMS-HSDPA Sytem" iley irele Communiation and Mobile Comuting Journal, 5, no. 4, : , June 005 [4] Aun Haider, and Rihard Harri, "A Note on the rformane of CP over HSDPA" ICICS, IEEE, : 5, 007. [5] Mohamad Aaad, and Djamal Zeghlahe," Cro-Layer Deign in HSDPA Sytem to Redue the CP Effet", IEEE Journal on eleted area in ommuniation, vol. 4, no. 3, : 64 65, Marh 006. [6] P eter arj an, and G abor N emeth, Buffer Overflow Probability of 0-4 CP Flow During Mobile Handover" IEEE Communiation Letter, vol., no. 7, : , July [7] Gábor Németh, Péter arján, Gergely Bizók, Feren Kubinzky, and Andrá Vere, " Meauring high-eed CP erformane during 0 0 mobile handover" 3nd IEEE Conferene on Loal Comuter 0 Network, IEEE Comuter oiety, : , [8] Majid Ghaderi, Ahwin Sridharan, and Hui Zang, "CP-Aware Channel Alloation in CDMA Network",IEEE ranation on Mobile Comuting, Vol. 8, No., : 4 8, Jan Normalized air time (% 00 0 [9] Eitan Altman et al., Analyi of the CP/IP flow ontrol mehanim Buffer ize (kbyte in high-eed wide-area network, in Pro. IEEE Conferene on Deiion and Control, : , 995. Figure 9. he overall Droing Probability v normalized air time η at [0] Mohamad Aaad, and Djamal Zeghlahe, Comarion between different Buffer Size for Rwired 50me. MIMO tehnique in a UMS HSDPA ytem, in Pro. IEEE Int. Sym. Sread Setrum eh. Al., Sydney, Autralia, , Se [] Chae D. "Code ombining A maximum-likelihood deoding aroah for ombining an arbitrary number of noiy aket" IEEE ranation on Communiation, : [] S. Floyd,. Henderon, and A. Gurtov, he NewReno Modifiation to CP Fat Reovery Algorithm, RFC 378, 004. IV. CONCLUSION AND FUURE ORK [3] Finaliing FRC referene lit and ulink ignalling definition, 3GPP inut aer SGR, 4(0, 00. hi aer derive an analytial model to determine the effet of CP rotool on the HSDPA ytem erformane. he HSDPA model wa onerned about the effet of having different buffer ize in the RNS ubytem. he reented analytial model wa derived to obtain the overall droing robability. he reult obtained are loe to the ublihed reult reorted in [7]. he model onfirm that the lower the ongetion rate of CP, the higher the aliation ervie rate. Additional inight on the HSDPA ytem behavior and interation with RNS buffer ize and CP RO i rovided. he reented work analyze the HSDPA and rooe a method to elet the aroriate ervie rate for having minimal overall droing robability. he reult howed that, dereaing the network ervie rate had a ignifiant imrovement on the overall dro robability. A mall hange in ervie rate enhane the overall ytem erformane by inreaing the utilization of air interfae. In other word, the network erformane will be more indeendent on the normalized air time a the differene between the ervie rate dereae. [4] 3GPP, Phyial layer aet of URA HSDPA, S V4.0.0, (Rel. 4. [5] S.M. Alamouti, A imle tranmit diverity tehnique for wirele ommuniation, Seleted Area in Communiation, IEEE Journal on, Volume: 6 Iue: 8, :45-458, Ot 998. [6] PRADEEP S. SINDHU, "Retranmiion Error Control with Memory", IEEE ranation on Communiation, Vol. om-5, no. 5, : , May 977. [7] G. G ómez, and R. S ánhez, End-to-End Quality of Servie over Cellular NetworkData Servie rformane and Otimization in G/3G, John iley & Son Ltd. England, 005. [8] Ioanni ara, and Vaili aouidi, "On the roertie of an adative CP Minimum RO" ProComuter Communiation Volume 3, Iue 5, Elevier, : , 7 Marh