Adaptive Packet Level Redundancy Mechanisms for Reliable Mobile Multicast: Proposed Architecture and Performance Analysis

Transcription

1 Adative Pacet Level edundanc echanisms for eliable obile ulticast: Proosed Architecture and Performance Analsis Prawit Chumchu *, **, and Aruna Seneviratne** ember, IEEE *Det. of Telecommunications Engineering ahanaorn Universit of Technolog, Bango Thailand **School of Electrical Engineering and Telecommunications The Universit of ew South Wales, Sdne Australia {chumchu, Abstract- In this aer, we investigate how acet level redundanc mechanisms could be combined with AQ (Automatic eeat request) to achieve scalable reliable mobile multicast transmissions. We consider three redundanc mechanisms. The first one called FEC-based mechanism is that redundant acets are achieved b FEC (Forward Error Correction) encoding. The second te called elicate-based mechanism is that redundant acets are achieved b relication. Combination of FECbased and relicate-based is used in the last mechanism called FEC-elicate-based mechanism. We also roose a novel adative acet level redundanc mechanism based on FEC-elicate-based mechanism. To design a novel acet level redundanc mechanism, we start with analzing comarative erformance of four reliable multicast mechanisms, which are Pure A-based mechanism, A-FEC-based mechanism, A-elicatebased mechanism, and A-FEC-elicate-based mechanism. In addition, simulation is rovided to evaluate the four mechanisms when both data acets and feed bac acets might be lost. The simulation and analsis show that FEC-elicate-based mechanism could be the best for using in reliable mobile multicast alications, which have different acet riorities. Index terms- obile ulticast, Adative redundanc, Secure multicast I. ITODUCTIO S CALABILITY should be strongl considered in reliable multicast []. To imrove scalable reliable multicast, combination of AQ (Automatic eeat equest) and acet level redundanc aroach obviousl decreases the number of reair acets leading to reduce A imlosion. evertheless, the roblem of setting the roer redundanc still remains. If the sender uses a fixed redundanc that is indeendent of the loss rates exerienced b receivers, then if the loss rates are much higher than what redundanc is able to recover losses, the sender ma still face from A imlosion. Also, the need for retransmission will increase the overall transmission latenc. On the other hand, if the loss rates are much lower than redicted, bandwidth is wasted due to the unnecessar redundant acets sent to receivers. In addition multicast members receive more dulicate acets leading to more ower rocessing time requirement. Consequentl, the adative FEC should be studied to adjust redundanc in reliable mobile multicast. A ure FEC redundanc mechanism, which adds uniform redundanc for each original acet in a FEC bloc, is not ICITA ISB: suitable in multicast alications with different acet riorities such as grou ree rotocol in secure multicast [] and distributed electronic stoc exchange. Therefore, relication could be used to rovide relicated acets for higher riorit original acets. The relication could reduce latenc in deliver because multicast receivers might receive the higher acet in the first round. FEC redundanc could be used to imrove scalabilit and reduce resonse acets from multicast receivers. In this aer we stud on adative acet level redundanc. Pacet level redundanc mechanisms could be classified in three mechanisms. The first one called FEC-based mechanism is that redundant acets are achieved b FEC (Forward Error Correction) encoding. The second te called elicate-based mechanism is that redundant acets are achieved b relication. Combination of FEC-based and elicate-based is used in the last mechanism called FEC- elicate-based mechanism. We start with comarative erformance analsis under the assumtion that A acets are never lost. Based on numerical results, we design a new adative acet level redundanc mechanism, which uses FEC-elicate-based mechanisms. Simulation is conducted to test the roosed mechanism under real situation. Simulation results and numerical results show that the roosed mechanism could control the average number of transmissions to be virtuall indeendent to the number of multicast members. The rest of the aer is structured as follows. In Section II, we rovide an overview of the related wors. Section III resents analsis and numerical results of acet level redundanc mechanisms. The roosed adative acet level mechanisms are resented in Section IV. The simulation results are resented in section V, and finall section VI concludes and summarizes our wors. II. ELATED WO In the literature, etzner [] and Huitema [] studied the use of hbrid AQ for reliable multi-oint transmission in the context of a bloc data transfer with indeendent and homogeneous loss. onnenmacher et al. [] extended the stud in the case of heterogeneous and burst loss. onnenmacher et al [] evaluated the use of FEC for reliable multicast. The classif using FEC into two tes: () laered FEC: () integrated FEC and AQ. In the first scheme, the reliable multicast laer and FEC laer are searated. In the second scheme uses retransmission of arit acets to recover the loss of original acets. All of the above aers do not tal about the detail of redundanc adatation mechanisms. iaein et al. [] have studied reliable multicast in wireless networs using adative

2 Pure FEC. The roosed theoretical basis for using adative FEC to rovide QoS guarantees in terms of reliabilit as well as efficienc of multicast deliver. ubenstein et al. [] investigated how to use roactive ure FEC to reduce feedbac imlosion and the exected dela of reliable deliver without increasing bandwidth usage. Authors in [] and [] studied the use of ure FEC for reliable multicast and did not design new adative mechanisms. Gemmell [7], izzo et al. [8], Yoon et al. [9] roosed fixed reliable multicast using adative ure FEC to imrove scalabilit. In [], cinle et al. imlemented reliable multicast b using roxies to suort collaboration among the mobile users. The utilized an adative ure FEC, which adjusts the level of redundanc in resonse to acet loss rate in the wireless networ. However these schemes do not resent the detail of analsis. III. AALYSIS COPAISO To roose a new adative redundanc mechanism, in this section we would resent comarative analsis of four reliable multicast mechanisms: Pure A-based mechanism (), A-FEC-based mechanism (), A-elicatebased mechanism () and A-FEC-elicate mechanism (). A. Analtical odel We assume that channel acet loss rate is mutuall indeendent. Also we assume that A acets are never lost. We analze mixture of multicasting and unicasting which is based on the maximum of number of transmissions via multicast, sa. Either ure multicasting or ure unicasting is not suitable for reliable mobile multicast since ure unicasting uses more bandwidth and ure multicasting maes more dulicate acets at receivers when number of receiver increases. The analsis focuses on reliable multicast deliver between a FA (Foreign Agent) or base station and Hs. The multicast acets deliver mechanisms are described as follow. echanism, : ulticast initialization Phase: The FA multicasts an original acet if a multicast receiver cannot successfull receive the acet, it sends a A acet to the source. ound = eair Phase: The FA checs whether the FA has received at least one A acet in the revious round. If es, it remulticasts the acet corresonding the A if is higher than the value of ound or unicasts the acet to each requesting receiver if is equal or less than the value of ound. If no, jum to Sto ound =ound+ Go to eair Phase Sto echanism,,, : ulticast initialization Phase: The FA multicasts a bloc of n acets consisting of original acets and h redundanc acets. The h redundanc acets consist of h acets generated from FEC ending using eed Solomon Erasure code and h acets generated b relication. When h= for, h= For and h and h are constants for If a multicast receiver has received x distinct acets, if x is less than it has to request -x original acets. The multicast receiver (H) sends a A acet for each requested acet. eair Phase: The FA checs whether the FA has received at least one A acet in the revious round. If es, it remulticasts the acet corresonding the A if is higher than the value of ound or reunicasts the acet to each requesting receiver if is equal and less than the value of ound. If no, jum to Sto ound =ound+ Go to eair Phase Sto Smbol eaning j Index of Hs ( Pacet loss rate at H j D E ( P Probabilit of needed FEC decoding Probabilit of needed FEC encoding E ) Average rocessing time requirement for receiving a data acet E ( P ) Average rocessing time requirement for receiving a A acet E ( P T ) Average rocessing time requirement for sending a A acet E ( P I ) Average rocessing time requirement for interrut E ( P DEFEC ) Average rocessing time requirement for FEC decoding a data acet E ( P EFEC ) Average rocessing time requirement for FEC encoding an original acet ρ Proactive factor = h/ Table : otation used in this aer B. Average Wireless Channel Utilization time and Processing Time equirement Analsis We start with analsis of average number of transmissions for an original acet E ( T ) and average number of receiving acets er original acet, E ( ). Then we analze the average number of A transmissions er original acet E ( T ), the average number of receiving A acets for an original acet E ( ), and the average number of interrut timer an original acet E I ). (

3 The robabilit that a acet needed request (q) Average number of Transmissions h=, h=. h=, h=. h=, h=. h=, h=. h=, h=. h=, h= simulation Analsis....8 Channel aet loss rate () Fig. The robabilit that a acet needed request h=, A-FEC. h=, A-FEC-e. h=, A-FEC-e. h=, A-FEC-e. Pure A umber of Hs Fig.. the average number of transmissions as a function of number of Hs Average number of Transmissions h=, A-FEC. h=, A-FEC-e. h=, A-FEC-e. h=, A-FEC-e. Pure A umber of Hs Fig.. The average number of transmissions as a function of number of Hs The average number of transmissions er original acet is T ) = + ρ + j= q( ( i= j= ( q( i ( )) +, ( ( ( ), when () where ρ is roactive factor, for, h/ for,, and. The value of q( is derived in the aendix The average number of received acets er original acet is ( ) = ( ( ) ( + ρ + + q( (, when ( i= j= ( q( i ( ))) The average number of A transmissions er original acet is T ( ) = q( ( ( ) /( ( ) + q( ( /( ( ) The average number of received A acets er original acet is ) = j= T ( ) The average number of timer interruts er original acets is I q( ( /( ( ), if = ( ) = q( ( ( ( ) /( ( ) + q( ( ( /( ( ), if B.. The average wireless channel utilization time er FA. Average wireless channel utilization time consists of the average channel utilization time for sending a data (original or redundanc) acet to be successful received b all multicast members (UDATA) and the average wireless channel utilization time for sending A acets er original acet (UA). UDATA is multicast acet size divided b wireless data rate and UA is multicast acet size divided b wireless data rate. The average wireless channel utilization time is given b: E ( U ) = UT ) + U A ) () where E U ) = UDATA( T ) (7) ( T and E U ) = UA ( T )). (8) ( A B.. Host rocessing time at each FA and each H. Firstl, consider the rocessing time at each FA. Average rocessing time requirement at each FA is comosed b the average rocessing time for sending a data acet to be successful received b all multicast members and the average rocessing time for receiving A acets er original acet. The average rocessing time at each FA, E ( P FA ), can be shown to be: E P ) = T ) P ) + ) P ) + P ). (9) ( FA T E EFEC We focus on the average rocessing time er original acet at a H. Average rocessing time requirement at a () () () ()

4 multicast receiver consists of the average rocessing time for receiving an original acet to be successful received b all multicast members, the average rocessing time for sending A acets er an original acet and the average time for rocessing A interrut timer for the original acet. Similar to (), we have PH ( ) = ( ) P ) + T ( ) PT ) + () I ( ) PI ) + D ( PDEFEC ). From (), it can be seen that if the value of q is controlled to be roortional to / ( q = / ) where is the number of Hs and is a constant which is average wireless channel acet loss rate. As a result the number of transmission is virtuall indeendent to the number of Hs. C. umerical results To understand the differences of the four generic mechanisms, we show the numerical results in this section. Firstl Figure is resented the robabilit q that a original acet needed to be requested. The channel acet lose rate was varied for to. It can be seen that o redundanc (.) mechanism roduces the highest q, which is the same as and FEC-based mechanism gives the lowest q. The q of FEC-e-based mechanism is in the range of q of o redundanc mechanism and FEC-based mechanism. Secondl we show the average number of transmissions er original acet in the case of indeendent and homogeneous channel acet loss rate and in the case of indeendent and heterogeneous channel acet loss rate. The number of FEC redundanc acets (h) was calculated as in Figure. Figure shows the average number of transmissions when homogeneous channel acet loss rate is.7. Figure resents the average number of transmissions when heterogeneous channel acet loss rates were randoml selected between.7 and.7. The number of original acets and maximum number of multicasting were set to be and resectivel. It can be seen that the average number of transmissions of A-FEC-based mechanism and A-FEC-elicate-based mechanism are virtuall indeendent to the number of Hs. In the case of A-FEC- elicate-based mechanism, the more relicated acets, the more the average number of transmissions. IV. POPOSED ADAPTIVE EDUDACY ECHAIS In this section, we would lie to resent how the roosed adative redundanc mechanism wors. Our adative mechanism is based on FEC-elicate-based mechanism. We divided the adative mechanism into arts: adative relicated redundanc and adative FEC redundanc. elicated redundanc acets are adated to riorities of original acets and FEC redundanc acets are adated to channel acet loss rate and number of multicast members. We start with introducing wireless loss arameter estimation. Then we show the adative FEC redundanc mechanism and the adative relicated redundanc mechanism. eceived a bloc of orignal acets and udate h acets n = +h, h= Calculate q( C < / Sto es no h=h+ Fig.. Adative FEC edundanc Algorithm. A. Wireless loss arameter Estimation Wireless loss model could be classified into classes: indeendent loss model and correlated loss model. The detail of wireless loss arameter estimation can be found in [, ]. In this aer we focus on indeendent loss model. Therefore onl acet loss rate,, is to be estimated and each H reorts the loss rate to its serving FA. The estimation of is erformed b calculating number of losses er a FEC bloc divided b the number of acets in the bloc. B. Adatation of FEC redundanc The adative FEC redundanc algorithm is shown in Fig.. It could be reresented as follows. Each FA receives wireless loss arameters. The criteria value C is calculated b averaging q ( for all Hs, C = q (. The j= value q ( is calculated as in (A) using the wireless channel acet loss rate. The h value is increased b one until the value of C is less than / where is average acet loss rate of all Hs and is the number of Hs. C. Adatation of relicated redundanc The relicated redundanc acets are adated to the different riorities of original acets in a FEC bloc. For instance, in a original acet FEC bloc, the first acets is much higher imortant than others. The adative relicated redundanc mechanism relicates the five acets. The relication could be one relicated acet for one original acet or / relicated acet for one original acet. The relication algorithm deends on the difference of original acet riorities. V. SIULATIO ESULTS In this section, we rovide a simulation to show comarative erformance in the real situation which both the data acets and feedbac acets are lost with bit error rate x -. S [] (etwor Simulation) is selected to simulate. The simulation scenario consists of one FA and to Hs.

5 Average number of transmissions Average Wireless Lin Utilization time (sec) Pure A. h=, A-FEC-elicate. A-FEC umber of Hs Fig.. The average number of transmissions er original acet as a function of number of Hs Pure A. h=, A-FEC-elicate. A-FEC umber of Hs Fig.. The average wireless channel utilization time as a function of number of Hs The Hs were randoml moved in the service area of the FA. The bit error rate is uncorrelated and all loss events at all Hs for all transmission are mutuall indeendent. Data acet size, and A size are btes, btes resectivel. IP and AC header size is 7 btes. The maximum number of multicasting was set to be one. The simulation results are shown in Fig.. Figure shows the average number of transmissions er original acet. It can be seen that the average number of transmissions of FEC-based is virtuall indeendent to the number of Hs and is significantl less than that of Pure- A-based mechanism. The wireless channel utilization time is shown in Fig.. It can be seen that the average wireless channel utilization of FEC-based is much less than that of Pure-A-based. Similar to the average number of transmissions, the wireless channel utilization time of FECbased is virtuall indeendent to number of Hs. VI. COCLUSIO We have studied three acet level redundanc mechanisms: FEC-based mechanism, elicate-based echanism and FEC-elicate-based mechanisms. The stud has been started from analsis under A acets are never lost. Analtical results are used to design a new adative acet level redundanc mechanism. Simulation is used to evaluate the roosed adative mechanisms in the real situation. Simulation results and numerical results show that the roosed mechanism could control the average number of transmissions to be virtuall indeendent to the number of multicast members. In addition, the roosed mechanism is suitable for the case of different riorit acets since the roosed mechanism rovides adative relicated redundanc. In future wor, we would design a new reliable grou ree transort rotocol based on A-FEC-elicate-based mechanism. APPEDI In this aendix we would resent the robabilit,,,, that a acet is lost after adding redundanc and the robabilit, q, q, q, q that a acet is needed to be requested after adding redundanc of all of the four acet redundanc mechanisms. In addition, the robabilit, D that a acet need to be FEC decoded and the robabilit, E that a acet need to be FEC encoded are shown the last art of this aendix. A. Analsis of mechanism In this mechanism, Because of no redundanc acets, the robabilities that a acet is lost and is requested are channel acet loss rate as follows: q ( = ( (A) and ( = q (. (A) B. Analsis of mechanism q n i i ( = ( i) ( ( ( ) i= n n ( = ( ( i= i C. Analsis of mechanism i ( ( ( ) n n i n i ) (A) (A) Firstl, the robabilit that acet in acets is needed to be requested q (, ) for H j is resented as follows: q j h ( ( ( ( )), uniform (, = h ( ( s x ( ( ( )), riorit x=. Thus (A) h ( ( ( ( )), uniform q = ( h ( ( s x ( ( ( )), riorit (A) = x= Where s x is the robabilit that original acet is selected to relicate at acet x of redundanc acets. Since lost acets and requested acets are the same,

6 ( = q (. (A7) D. Analsis of mechanism In this aendix we show the calculation of the robabilit of x losses in a sequence of acets, P (,. The calculation of, is following:, = ( ( ), + ( x,,, ( ( ), + ( x,,, < + h (A8) For =,,... + h and x =,,,...,. The variable h is the number of redundanc acets generated b FEC encoding and ( is loss robabilit of acet of H j which could be calculated as: h ( = ( ( s x ( ( ( )) (A9) x= where s x is the robabilit of acet is selected to add redundanc acet at acet x in h relicated acets b relication. The initial conditions for recursion are P (,) =, ) =, x. (A) Accordingl, the robabilit of a acet requested after FEC decoding, q, is + h q ( = ( x h) + h, (A) x= h+ ext we show the calculation of erceived acet loss rate of acet, (, ), as following j n i= (, = PE ( i, n, ( (A) Where n = +h and P E ( i, n ) is the robabilit of i losses in a sequence of n acets, which do not include acet. D, for n (( ( + ( n ( ( ) ( ), for, for ( = h ( ( ( ) + ( = h ( ( ( )), for = (A) E ( = for, for, for and for (A) EFEECES [] A. anin, A. omanow, S. Bradner, and V. Paxson IETF Criteria for Evaluating eliable ulticast Transort and Alication Protocols, FC 7, Informational, June 998. [] J. etzner, An imroved broadcast retransmission rotocol, IEEE Transactions on Communications, CO-():79 8, June 98. [] C. Huitema, The case for acet level fec, In Proceedings of IFIP th International Wor-shoon Protocols for High Seed etwors (PfHS 9), IIA, Sohia Antiolis, FACE, October 99. [] J. onnenmacher, E. W. Biersac, and D. Towsle, Parit-based loss recover for reliable multicast transmission etworing, IEEE/AC Transactions on, Volume: Issue:, Aug. 998 Page(s): 9. []. iaein, H. Labiod and C. Bonnet, A-FEC: A QoS-Besed Adative FEC for ulticast Communication in Wireless etwors, Proc IEEE,. 9-98,. [] D. ubenstein, J. urose, and D. Towsle, A Stud of Proactive Hbrid FEC/AQ and Scalable Feedbac Techniques for eliable eal-time ulticast, Comuter Communications Journal, arch. [7] J. Gemmell, Scalable eliable ulticast Using Erasure-Correcting e-sends, June 997, icrosoft esearch Technical eort, S-T-97-, June 997. [8] L. izzo and L. Vicisano, DP: an FEC-based eliable ulticast rotocol for wireless environments, AC obile Comuting and Communications eview, Vol.,., Aril 998. [9] J. Yoon, A. Bestavros, and I. atta, Adative eliable ulticast (A), in the International Conference on Communications (ICC), ew Orleans, Louisiana, June 8-,. [] P.. cinle and A.P. ani, An exerimental stud of adative forward error correction for wireless collaborative comuting, Alications and the Internet,. Proceedings. Smosium on,. 7. [] P. Chumchu, Z. G. Zhou and A. Seneviratne, An Analsis Comarison of eliabilit echanisms for Imroving eliable obile ulticast Protocol, submitted to IEEE Trans on mobile comuting. [] Chumchu, Z. G. Zhou and A. Seneviratne, A Scalable eliable obile ulticast Protocol, submitted to IEEE Trans on mobile comuting. [] Chung ei Wong, ohamed G. Gouda, and Simon S. Lam, Secure Grou Communications Using e Grahs, ' IEEE/AC Transactions on etworing, Feb.. [] UCB/LBL/VIT etwor Simulator (S), htt://