Design of a prioritized error control scheme based on load differentiation for time sensitive traffic on the wireless LAN

Transcription

1 JOURNAL O NETWORKS, VOL. 1, NO. 2, JUNE Desgn of a prortzed error control scheme based on load dfferentaton for tme senstve traffc on the wreless LAN Mkyung Kang, Junghoon Lee*, Yongmoon Jn, Gyung-Leen Park, Hanl Km Dept. of Computer Scence and Statstcs, Cheju Natonal Unversty, , Jeju Do, Republc of Korea Emal: {mkkang, jhlee, ymjn, glpark, hkm}@cheju.ac.kr Abstract Ths paper proposes and analyzes the performance of a prortzed error control scheme for tme senstve applcaton on the wreless sensor network. As a modfed verson of IEEE WLAN, the proposed scheme further dvdes DC nto H-DC and L-DC wthout changng PC, amng at maxmzng the successful retransmsson of a packet that carres crtcal data. Whle channel estmaton elmnates the unnecessary polls to the sensor node currently unreachable durng PC, two DC subperods enable prortzed error recovery by makng only the hgh prorty packet be retransmtted va H-DC. A good chop value, whch dstrbutes the retransmsson to each perod, can maxmze recovered weght, or crtcalty, mnmzng the possble degradaton of network throughput. The smulaton results show that the proposed scheme can mprove recovered weght by 8% whle showng 97% successful transmsson at maxmum for the gven smulaton parameter. Index Terms Tme-senstve applcaton, wreless sensor network, IEEE WLAN, Prortzed error control, Packet crtcalty I. INTRODUCTION In the past few years, smart sensor devces have matured to the pont that t s now feasble to deploy a large, dstrbuted network of such sensors [1]. Sensor networks are dfferentated from other wreless, batterypowered envronments n that they consst of tens or hundreds of autonomous nodes that operate wthout human nteractons for a long tme. The sensor network makes possble a new range of applcaton such as envronmental montorng and control. Wreless sensor/actuator networks allow scentsts to montor remote envronmental condtons such as temperature, pressure, and chemcal presence. Such montorng network can be used to detect forest fres and alert authortes, or even extngush the fre. Besdes, the Based on Desgn of an effcent error control scheme for tmesenstve applcaton on the wreless sensor network based on IEEE standard, by Junghoon Lee, Mkyung Kang, Yongmoon Jn, Gyungleen Park, and Hanl Km whch appeared n the Proceedngs (LNCS) on IWDC 2005, Kharagpur, Inda, December * Correspondng author applcaton of sensor networks s beng expanded to ndustral, mltary, agrcultural, and other varous areas. Communcaton between the sensors and snks requres a network [2]. Snce sensor applcatons preclude the use of wred networks, wreless networks are commonly used n those applcatons. Wreless networks are nherently broadcast meda, and all nodes n the network share one common communcaton medum. Therefore, a method for resolvng contenton when multple nodes requre access to the medum s necessary. Ths s the purpose of a MAC (Medum Access Control) protocol. The MAC protocol defnes how and when nodes may access the medum. It must ensure that nodes share the medum n such a way that applcaton requrements are met. Message flows exchanged n a sensor network are manly perodc and need guaranteed delay for a computng node to make a meanngful and tmely decson [3]. Many realtme schedulng and far packet schedulng algorthms have been developed for wred networks, however, t s not clear how well these algorthms work for wreless sensor networks where channels are subject to unpredctable locaton-dependent and tme-varyng bursty errors [4]. The wreless lnk s hghly varable even over short dstances due to the statstcal dstrbuton of path loss and the physcal propertes of propagaton envronment. In the presence of such unpredctable errors, real-tme traffc applcatons cannot fully utlze the channel bandwdth assgned to them. The IEEE was developed as a MAC standard for WLAN [5]. The standard conssts of a basc DC (Dstrbuted Coordnaton uncton) and an optonal PC (Pont Coordnaton uncton). The DC explots CSMA/CA (Carrer Sense Multple Access wth Collson Avodance) protocol for non-real-tme messages, amng at enhancng ther average delvery tme as well as network utlzaton. However, packet collsons, ntrnsc to CSMA protocols, make t mpossble for a node to predctably access the network. After all, QoS guarantee cannot be provded wthout developng a determnstc access schedule on top of collson-free PC, n whch a pont coordnator polls actve sensor nodes [6]. The pollng schedule s decded accordng to the contract negotated between each flow and the pont coordnator.

2 46 JOURNAL O NETWORKS, VOL. 1, NO. 2, JUNE 2006 PC thus makes WLAN as a powerful and promsng nfrastructure for the wreless sensor network of tmesenstve sensor applcaton. However, WLAN needs to effcently deal wth network error durng the delvery of sensory data. rst, t s desrable to skp pollng a node whose channel s not n normal condton. Second, once the packet transmsson fals, t should be retred n a best-effort manner wthn ts deadlne. Thrd, due to the bursty nature of wreless network error, not all packets can be recovered by retransmsson. In that case, the retransmsson should carefully consder the prorty of a packet, and network should try to enhance the successful retransmsson of hgher prorty packets. The prorty of packet can be decded by the mportance of data the packet carres, or by the degree of QoS degradaton the flow experenced [4]. If we consder a montorng applcaton, a momentary event data s more mportant than other steady state data. The bandwdth for the respectve error control packet cannot be reserved for each flow durng PC, for t wll lead to the great waste of network bandwdth and neffcency. In addton, t s mpossble for AP to know whch node wants to be polled for retransmsson on a specfc tme. Hence, t s natural that error control s performed on DC, makng sensor nodes farly share the bandwdth. Implementng a global prorty resoluton n DC seems to be unrealstc as t needs major modfcaton on the orgnal CSMA/CA protocol as well as prorty resolvng procedure for every packet transmsson nvokes so much overhead that can be avoded when there s no contendng message [7]. In ths paper, we propose and analyze an error control scheme for sensor data on DC nterval of IEEE WLAN. It s amng at supportng, though lmted, level of prorty n recoverng the packet transmsson error, durng DC wthout greatly modfyng the orgnal CSMA/CA protocol. To ths end, AP dvdes the DC nto two subperods, makes ther loads dfferent, and gves more chance to the hgher prorty message by transmttng t va lower load network. Ths polcy enhances the possblty of successful transmsson for hgher prorty packets, nsgnfcantly sacrfcng the overall network throughput. The rest of ths paper s organzed as follows: Secton II ntroduces the background of ths paper, ncludng the related works on real-tme communcatons on the wreless medum, IEEE Wreless LAN standards, and real-tme message model. Then Secton III proposes the communcaton archtecture for tme-senstve sensor traffc, descrbng channel estmaton, bandwdth allocaton, and prortzed error control. After demonstratng the smulaton result n Secton IV, Secton V concludes ths paper wth a bref summarzaton and the descrpton of future works. II. BACKGROUND A. Related Works As an example of pure bandwdth allocaton scheme, DBASE (Dstrbuted Bandwdth Allocaton/ Sharng/ Extenson) s a protocol capable of supportng both synchronous and multmeda traffcs over IEEE ad hoc WLAN where no fxed access pont coordnates medum access [8]. The basc concept s that each tme real-tme staton transmts ts packet t wll also declare and reserve the needed bandwdth at the next CP. Every staton collects ths nformaton and then calculates ts actual bandwdth at the next cycle. But t focuses on decdng whether to accept the respectve real-tme packet wthout consderng a flow-level allocaton or negotaton. M. Caccamo and et. al. have proposed a MAC that supports determnstc real-tme schedulng va the mplementaton of TDMA (Tme Dvson Multple Access), n whch the tme axs s dvded nto fxed sze slots [6]. Referred as mplct contenton, ther scheme makes every staton concurrently run the common realtme schedulng algorthm to determne whch message can access the medum. Each message mplctly contends for the medum through the schedulng algorthm, for example wth prortes, nstead of explctly on the medum. Unfortunately, to mplement mplct contenton, each node must schedule all messages n the network, resultng n the complexty growng lnearly wth the number of messages. In addton, ther scheme ddn't consder the network error at all, just focusng on the schedulng polcy. Cho and Shn suggested a unfed protocol for realtme and non-real-tme communcatons n wreless networks [3]. In ther scheme, a BS (Base Staton) polls a real-tme moble staton accordng to the non-preemptve ED (Earlest Deadlne rst) polcy. The BS also polls the non-real-tme message accordng to the modfed round-robn scheme regardless of a standard CSMA/CA protocol to completely elmnate a message collson. The retransmsson of a damaged packet s consdered as a normal non-real-tme message. Addtonally, to handle locaton-dependent, tme-varyng, and bursty channel errors, the channel state can be predcted va channel probng before the packet s transmtted. That s, before transmttng a downlnk packet to the moble, the BS transmts a probng control packet to that moble, whch then returns the control packet to the BS. Wth ths estmaton, t s possble to reduce the need for retransmsson, snce retransmsson can be harmful n meetng deadlnes. Adamou and hs colleagues have addressed the schedulng problem of achevng farness among realtme flows wth deadlne constrants as well as maxmzng the throughput of all the real-tme flows over a wreless LAN [4]. They chose the schedulng objectve of mnmzng the maxmum degree of the degraded QoS among all applcatons. Ther schedulng polcy ncludes ED (Earlest Deadlne rst), GD (Greatest Degradaton rst), EOG (ED Or GD), and L (Laggng lows rst). The BS performs the schedulng of real-tme packet delveres usng a pollng scheme. Ths scheme s bult on the assumpton that BS knows

3 JOURNAL O NETWORKS, VOL. 1, NO. 2, JUNE whch staton has messages to retransmt as well as ther deadlnes, and decdes whch one to poll among them accordng to the crtera descrbed above. In addton to the retransmsson technque, there exsts a way the sender adjusts the transmsson rate accordng to the currently avalable network bandwdth. Shah proposed a dynamc bandwdth management scheme n a sngle-hop ad hoc wreless network that s especally sutable for hot-spot networks, that s, a number of nodes n a small area share lmted channel bandwdth [9]. As the avalable channel capacty changes and also the traffc characterstcs of varous flows change, the bandwdth manager dynamcally reallocates the channel access tme to the ndvdual flows. When network contenton and data rates are low due to the poor network condton, the transmsson queue can be draned faster than samplng results arrve [1]. However, because the number of messages produced durng a sngle epoch can vary dramatcally, there are stuatons when the queue wll overflow. In ths case, the system must decde whch one to dscard. In the nave scheme, no data s consdered more valuable than any other, so the queue s draned n a IO manner and data are dropped f they do not ft n the queue. The wnavg scheme works smlarly, except that nstead of droppng results when the queue flls, the two results at the head of the queue are averaged to make room for the new result. Snce the head of the queue s now an average of multple records, we assocate a count wth t. In the delta scheme, a data value s assgned an ntal score relatve to ts dfference from the most recent value successfully transmtted from ths node, and at each pont n tme, the data wth the hghest score wll be delvered. B. IEEE WLAN Two dfferent key approaches can be followed n the mplementaton of a WLAN: an nfrastructure-based approach and an ad-hoc networkng one. The nfrastructure-based archtecture mposes the exstence of a centralzed controller for each cell, often referred to as AP (Access Pont). The AP s normally a gateway to the external networks, for example, wred backbone network, or other wreless networks. Whereas, an ad-hoc network s a peer-to-peer network formed by a set of statons wthn the range of each other that dynamcally confgure themselves to set up a temporary network. PC access cannot be adopted n ad-hoc networks. The wreless LAN operates on both CP (Collson Perod) and CP (Collson ree Perod) phases alternatvely n BSS (Basc Servce Set) as shown n g. 1. Each superframe conssts of CP and CP, whch are mapped to PC and DC, respectvely. PC (Pont Coordnator) node, typcally AP, sequentally polls each staton durng CP. Even n the ad hoc mode, t s possble to desgnate a specfc node to play a role of PC n a target group. Only the polled node s gven the rght to transmt ts message for a predefned tme nterval, and t always responds to a poll mmedately whether t has a pendng message or not. In contrast, DC s the bass of the standard CSMA/CA access mechansm and t uses the RTS (Request To Send)/CTS (Clear To Send) clearng technque to further reduce the possblty of collsons. After sendng a RTS, subsequent steps take place usng SIS (Short Interrame Space), so the transmsson of a frame s an atomc operaton that cannot be nterrupted by another packet. The phase of network operaton s managed by the exchange of control packets whch have hgher prorty than other packets. The prortzed access s acheved by dfferent length of IS the node wats before t attempts to send ts packet. The PC attempts to ntate CP by broadcastng a Beacon at regular ntervals derved from a network parameter of CPRate. Round robn s one of the popular pollng polces for CP, n whch every node s polled once a pollng round. A pollng round may be completed wthn one superframe, or spreads over more than one superframe. In case the CP termnates before all statons have been completely polled, the pollng lst s resumed at the next node n the ensung CP cycle. Start CP Poll Ack CP (PC) End CP CP (DC) gure 1. Tme axs of wreless LAN Tme Start CP g. 1 also llustrates that poll, transmsson, and acknowledgment are atomc, namely, these steps must complete n ther entrety to be successful. Senders expect acknowledgment for each transmtted frame and are responsble for retryng the transmsson. After all, error detecton and recovery s up to the sender staton, as postve acknowledgments are the only ndcaton of success. If an acknowledgment s expected but does not arrve, the sender consders the transmsson faled. C. Message Model There s no other traffc but the sensor data n the network. The traffc of sensored data s typcally sochronous (or synchronous), consstng of message streams that are generated by ther sources on a contnung bass and delvered to ther respectve destnatons also on a contnung bass [10]. A node, currently nactve, can be actvated by an upper layer query command that wants to montor or process the data flow from the sensor node. The destnaton of message can be ether wthn a cell or outsde a cell, and the outbound messages are frst sent to the router node such as AP and then forwarded to the fnal destnaton. Internal messages are also relayed by the AP. The query may also specfy samplng perod and the precson level of needed data (hence, message length) on a node. In case of a change n the actve flow set, bandwdth s to be reallocated or network schedule mode s changed [9]. Ths paper follows the general real-tme message model whch has n streams, namely, S 1, S 2,..., S n, and for each S, a message szed less than C s produced at the begnnng of ts perod, P. Each packet must be delvered to ts destnaton wthn D unts of tme from ts

4 48 JOURNAL O NETWORKS, VOL. 1, NO. 2, JUNE 2006 generaton or arrval at the source, otherwse, the packet s consdered to be lost. We assume that D s larger than P to gve a suffcent margn to recover transmsson falures. When S s polled, t can transmt up to H, and {H } s named as capacty vector. The sampled data has ts own weght and the weght should be mapped to the correspondng prorty level. III. MESSAGE SCHEDULING SCHEME A. Overvew of Network Operaton The entre network conssts of a number of autonomous cells and each of them has ts own AP to manage the message schedule. After a query ssued to the entre sensor network s analyzed and dssemnated, each subquery s assgned to respectve network to actvate a necessary sensor operaton. Hence, we focus on the bandwdth allocaton and error control wthn a cell. Accordng to the operaton of AP, the tme axs of WLAN conssts of a seres of superframes and each of them conssts of PC, H-DC, and L-DC, vrtually dvdng the lnk nto 3 dscrete channels. As prevously mentoned, the beacon frame ntates each phase one by one. Naturally, each channel can nterfere wth one another, due to the deferred beacon problem, that s, a beacon message can get delayed and the start of PC can be put off, f another packet s already occupyng the network. Ths s because only after the medum s dle the coordnator wll get the prorty through the shorter IS. The maxmum amount of deferment concdes wth the maxmum length of a data packet, as can be nferred n g. 2. Ths nterference may jeopardze the sophstcated network schedule that guarantees satsfyng QoS requrement, partcularly n case the deadlne of message s equal to ts perod. Otherwse, the delayed start of PC just nduces a delay jtter that can be absorbed by the recever. Addtonally, we assume that the length of PC and that of H-DC are not reduced even f ther starts are delayed. Just L-DC shrnks ts length when ts start gets delayed, as shown n the rght-hand part of g. 2. After all, the tme partton s equvalent to the three ndependent lnks. Superframe PC H-DC L-DC PC H-DC L-DC PC H-DC L-DC PC H-DC L-DC Beacon DC Stretch Deferred Beacon gure 2. Tme axs of proposed network Tme Guaranteed flows occupy the PC under the exclusve control of AP. Each node transmts ts message on each poll for the predefned tme duraton decded by a specfc bandwdth allocaton scheme. AP polls only those nodes whose channel s estmated to be good. Hence, durng PC, AP does not poll a node f ts channel status s not estmated as good, snce t has almost no possblty to success consderng the error characterstcs of wreless channel. If a transmsson fals or s deferred, the sender moves the packet to the retransmsson queue, as ts mportance s not as hgh as newly generated sensor data. The packet may have a chance to be retransmtted at that queue va H-DC or L-DC accordng to ts prorty. B. Channel Estmaton The rado channel s modeled as a Glbert channel [11]. We can denote the transton probablty from state good to state bad by p and the probablty from state bad to state good by q. The par of p and q representng a range of channel condtons, has been obtaned by usng the trace-based channel estmaton. The average error probablty, denoted by ε, and the average p p q length of a burst of errors are derved as + and 1 q, respectvely. The packet s receved correctly f the channel s n state good for the whole duraton of packet transmsson, otherwse, t s receved n error. In case of an unsuccessful transmsson, the staton may retransmt the frame through ether another CP or CP accordng to the control polcy. We take the estmaton method from Bottglengo's work [12]. To trace the channel status, AP mantans a state machne, or smply flag, assocated to each sensor node. The state can be ether good or bad accordng to the channel condton. The channel condton s estmated as follows: The ACK/NAK s sent from the recever to AP as soon as t receves a packet. If the AP does not receve an ACK/NAK wthn predefned tme-out nterval, the packet wll be assumed to be lost. AP sets the state to good whenever t receves from the correspondng sensor node, namely, a MAC-layer acknowledgment n response to a data frame, a CTS frame n response to an RTS frame, or any other error-free frame. The AP sets the state to bad after a transmsson falure. Each bad channel has ts own counter, and when a counter expres the AP attempts to send a sngle data frame to check the channel status. The duraton of tmer s reset to ts ntal value upon a transmsson from bad to good, and the value s doubled whenever the probng fals n bad state. The value of tmer should be set small so as to quckly recover from short channel error perod. C. Bandwdth Allocaton By allocaton, we mean the procedure of determnng capacty vector, {H }, for the gven superframe tme,, as well as message stream set, {S (P, C )}. Ths step s necessary for the network to guarantee the tmely delvery of data packet generated by currently actve sensors. or each change n the set of actve sensors or ther traffc characterstcs, the pollng schedule and capacty vector should be reallocated. The samplng perod and the message length are dfferent for each sensor, or they can be assumed to be tunable. There have been a lot of bandwdth allocaton schemes for the realtme message stream or sensor data stream, we begn wth Lee's scheme for ts message model concdes wth that of ths paper. However, any other bandwdth allocaton

5 JOURNAL O NETWORKS, VOL. 1, NO. 2, JUNE scheme can be appled wth a trval modfcaton of ther basc assumpton or network model [13]. It s desrable that the superframe tme s a hyperperod of each stream's perod and t s known that a message set can be made harmonc by reducng some perods by at most half [2]. So we assume that the superframe tme s also gven n advance, focusng on the determnaton of capacty vector. H lmts the maxmum tme amount for whch S can send ts message. Let δ denote the total overhead of a superframe ncludng pollng latency, IS and the lke, whle D max the maxmum length of a data packet. or a mnmal requrement, should be suffcently large enough to make the pollng overhead nsgnfcant. In addton, f P mn s the smallest element of set P, should be less than P mn so that every stream can meet at least one superframe wthn ts perod. After all, the requrement for the superframe tme,, can be summarzed as follows: H δ + D P (1) + max mn In addton, the mnmum value of avalable transmsson tme, X s calculated as n (2). Namely, X ( ) ( ) P P = 1 H f P Dmax (2) X P = H Otherwse or each message stream, X should be greater than or equal to C (X C ). By substtutng (2) for ths nequalty, we can obtan the least bound of H that can meet the tme constrant of S. C H ( ) ( ) P = P f P D 1 max (3) C H = Otherwse P The allocaton vector calculated by (3) s a feasble schedule f t satsfes (1). By ths, we can determne the length of CP perod (T CP ) and that of CP (T CP ) as follows: TCP = H + δ, TCP = TCP D (4) max D. Schedulng of Retransmsson As shown n g. 3, each sensor node has two separate queues, namely, normal queue for ordnary packets and retransmsson queue for the packets that are to be resent. Packets are ordered accordng to ther prorty n the retransmsson queue, and a packet entry s automatcally dropped when ts deadlne expres. Ths fgure also llustrates that the proposed system has 3 vrtual transmsson lnks, PC lnk, hgh-prorty DC lnk, and low-prorty DC lnk, whle each of them s mapped to PC, H-DC, and L-DC perods, respectvely. If a channel between AP and a node has been n a bad state for not a short duraton, the number of packets to retransmt also ncreases. In that case, not all of packets can be recovered va retransmsson due to the lmt of avalable bandwdth. So t s necessary to gve packets a certan type of prorty and the network should gve the precedence to the hgher prorty packets. The prorty s typcally decded by the mportance of the value, or the degree of qualty degradaton for a flow. Packet Error Normal Queue Retransmsson Queue gure 3. Queue Dscplne Prorty Dstrbuton (H,L)-DC Wmn c PC Lnk Hgh Prorty DC Lnk Low Prorty DC Lnk The regular packets are transmtted va PC lnk when AP polls a node, and the pollng schedule s decded by the bandwdth allocaton scheme. In case AP skps some nodes as they are n bad state, other nodes can be polled more than guaranteed. Besdes the node possbly has no packet n ts normal queue, then t sends a packet n the retransmsson queue, f any. The operaton of H-DC and L-DC s as follows: The lower the load, the hgher the probablty of successful transmsson. Hence, we are to make the load of H-DC lower than that of L-DC, actually dfferentatng the upper bounds of maxmum load for two perods. However, as there s no global vew for each node, they cannot know whether other nodes have hgher prorty packets or not. Consequently, H- DC transmts those packets whose prorty s hgher than c, as shown n g. 4. Namely, they do not permt low prorty packet to be resent durng H-DC, even f there s no traffc n H-DC. It s possble for some packets to collde due to the same backoff value. Then the transmsson wll be retred after another backoff procedure just wthn one H-DC. If a packet recovery fals n H-DC, t can be retred n the L-DC wth a normal CSMA/CA procedure. H-DC gure 4. Chop Partton Wmax The value c s a tunable parameter that can be set accordng to the network load, current error rate, weght dstrbuton, and so on [14]. It ranges from the lowest prorty value, W mn to the hghest one, W max. If c s set to W max, t does not use the prortzed retransmsson. The optmal value of c whch maxmzes value of recovered weght, can be found emprcally or va analytcal model for the gven network parameters. Otherwse, some functons lke bandwdth management may dynamcally adjust the chop value. However, f a value exsts and once a value s found, t can be used or adjusted accordng to the change of network parameter. Hence t s our concern to confrm that such values exst. It s natural that the

6 50 JOURNAL O NETWORKS, VOL. 1, NO. 2, JUNE 2006 number of recovered packet wth non-parttoned scheme s obvously larger than that wth the proposed parttonng scheme, as the latter leaves the H-DC vacant to always reserve bandwdth for the hgher prorty packets, resultng n declnng network throughput "ProposedScheme" "NonParttoned" IV. PERORMANCE ANALYSIS At frst, we consdered ns-2 whch s the most common smulaton tool for wreless network, but t lacks the support for drect lnk layer control or error smulaton model [15]. Thus the experments are fulflled va SMPL [16], whch s functonally equvalent to ns-2 event scheduler. Wth SMPL, we mplemented restrcted contenton protocol based on RTS/CTS mechansm for DC. To concentrate on the prortzed error recovery n wreless sensor networks, we smplfed the experment as follows: rst of all, all tme varables are algned to the length of superframe tme,. Every stream has equal perod, message length, and deadlne, and exactly dvdes P whle the deadlne s fxed to 5. Each packet s generated at constant ntervals wthout any jtter and sent as a frame. The number of actve sensors s 5 and ther utlzaton s 0.5, whle capacty vector s {0.1, 0.1,..., 0.1}. Each packet fts to the length of 0.1, beng assocated to a prorty randomly pcked from 0 to 19. We also, rather unrealstcally, assume that error estmaton s always correct. The frst experment measures the effect of chop value wth fxed error rate, ε. The ε s set to 0.01, whle the length of error duraton, denoted as q 1 n Glbert error model, dstrbutes exponentally wth average 2.0. The y-axs of g. 5 s the rato of total weghts of recovered packets to those of packets that need retransmsson. As natural, some packets are dscarded at the retransmsson queue due to deadlne expraton. As shown n g. 5, the chop value s crtcal to the overall performance. Ths fgure also plots the recovered weghts by retransmsson va non-parttoned DC through ordnary CSMA/CA protocol to compare wth the proposed scheme. The gap between the two curves s maxmzed when chop value s After that pont, the larger s chop value, the more bandwdth s wasted. As contrast, wth smaller chop value, most of packets are attempted on H-DC and agan on L-DC, ncreasng the possblty of collson. g. 6 plots the measurement result of recovered weghts accordng to the ε rangng from 10-3 to or each ε value, each experment runs wth ts own chop value, and then recovered weght wth optmal chop value s pcked to plot the curve. As shown n the fgure, the proposed scheme always outperforms the nonparttoned retransmsson and acheves almost 97% of success of transmsson for the gven network and error parameter. V. CONCLUSION In ths paper, we have proposed and analyzed the performance of communcaton archtecture capable of effcently dealng wth channel error on the wreless Recovered Weght Weght Rato Chop Value gure 5. Recovered weght vs. chop value "ProposedScheme" "NonParttoned" Error Rate gure 6. Total weght vs. error rate sensor network for the tme-senstve sensor applcaton based on the IEEE Wreless LAN standard. The proposed scheme makes AP always estmate channel status between tself and each sensor node, to avod pollng a node whose channel s not n normal condton. Once the packet transmsson fals, t should be retred n a best-effort manner wthn ts deadlne. nally, t can support the prortzed error recovery by dvdng the DC nto two subperods and dfferentatng ther load. The experment performed va smulaton usng SMPL shows that the proposed scheme, wth ths DC partton, can mprove the recovered weght compared wth the non-parttoned scheme f a good chop value s found. or the gven envronment parameters, t shows about 8 % mprovement when the chop value s In addton, as for the sum of weghts of successfully transmtted packets, the proposed scheme always outperforms non-parttoned scheme. As a future work, we are to scrutnze the method to fnd the optmal chop value for the gven mportance dstrbuton as well as other real-tme communcaton parameters. Addtonally, the work that combnes the proposed communcaton archtecture, wll target to tme-

7 JOURNAL O NETWORKS, VOL. 1, NO. 2, JUNE senstve sensor applcaton, wth power management schemes. [16] MacDougall, M., Smulatng Computer Systems: Technques and Tools, MIT Press, ACKNOWLEDGMENT Ths research was supported by the MIC (Mnstry of Informaton and Communcaton), Korea, under the ITRC (Informaton Technology Research Center) support program supervsed by the IITA (Insttute of Informaton Technology Assessment) (IITA-2005-C ). REERENCES [1] Madden, S., rankln, M., Hellersten, J., Hong, W., The desgn of an acqustonal query processor for sensor networks, ACM SINGMOD, [2] Carley, T., Ba, M., Barua, R., Stewart, D., Contentonfree perodc message scheduler medum access control n wreless sensor/actuator networks, Proc. IEEE Real-Tme Systems Symposum, Dec [3] Cho, S., Shn, K., A unfed wreless LAN archtecture for real-tme and non-real-tme communcaton servces, IEEE/ACM Trans. on Networkng, pp.44-59, eb [4] Adamou, M., Khanna, S., Lee, I., Shn, I., Zhou, S., ar real-tme traffc schedulng over a wreless LAN, Proc. IEEE Real-Tme Systems Symposum, pp , Dec [5] IEEE , Part 11: Wreless LAN Medum Access Control (MAC) and Physcal Layer (PHY) Specfcatons, also avalable at [6] Caccamo, M., Zhang, L., Sha, L., Buttazzo, G., An mplct prortzed access protocol for wreless sensor networks, Proc. IEEE Real-Tme Systems Symposum, Dec [7] Vadya, N., Bahl, P., Gupta, S., Dstrbuted far schedulng n a wreless LAN, 6-th Annual Int'l Conference on Moble Computng and Networkng, Aug [8] Sheu, S., Sheu, T., A bandwdth allocaton/sharng/ extenson protocol for multmeda over IEEE ad hoc wreless LANS, IEEE Journal on Selected Areas n Communcatons, Vol. 19, No. 10, pp , Oct [9] Shah, S. H., Chen, K., Nahrstedt, K., Dynamc Bandwdth Management for Sngle-hop Ad Hoc Wreless Networks, ACM/Kluwer Moble Networks and Applcatons (MONET) Journal. 10, pp , [10] Lu, J., Real-Tme Systems, Prentce Hall, [11] Ba, H., Atquzzaman, M., Error modelng schemes for fadng channels n wreless communcatons: A survey, IEEE Communcatons Surveys, Vol. 5, No. 2, pp.2-9, [12] Bottglengo, M., Casett, C., Chasern, C., Meo, M., Short term farness for TCP flows n b WLANs, Proc. IEEE INOCOM, [13] Lee, J., Kang, M., Jn, Y., Km, H., Km, J., An effcent bandwdth management scheme for a hard real-tme fuzzy control system based on the wreless LAN, accepted to LNCS: Embedded Systems for Ubqutous Computng, [14] Gao, B., Garca-Molna, H., Schedulng soft real-tme jobs over dual non-real-tme servers, IEEE Trans. Parallel and Dstrbuted Systems, pp.56-68, Jan [15] all, K., Varadhan, K., Ns notes and documentaton, Techncal Report, VINT project, UC-Berkeley and LBNL, Mkyung Kang receved the B.S. and M.S. degrees at Dept. of computer scence and statstcs from Cheju Natonal Unversty, Jeju, Korea. Currently, she s a Ph.D. canddate at Cheju Natonal Unversty, Korea. Her research nterests are real-tme communcaton and wreless network. She s a researcher of Key Technologes for telematcs systems at ITRC (Informaton Technology Research Center) of Cheju Natonal Unversty, Korea. Junghoon Lee receved the B.S., M.S., and Ph.D. degrees at Dept. computer engneerng, Seoul Natonal Unversty, Korea. Hs research nterests nclude real-tme communcaton and wreless network. rom 1990 to 1992, and also n 1996, he was a senor research engneer at Lab. of optcal telecommuncaton, Mercury, Korea. In 1997, he joned the Dept. of computer scence and statstcs at Cheju Natonal Unversty, Korea where he s currently an assocate professor. rom 2003 to 2005, he was a vstng scholar at Dept. of computer scence, Unversty of Texas at Austn. Yongmoon Jn receved the B.S. and M.S. degrees at Dept. of computer scence and statstcs, Cheju Natonal Unversty, Jeju, Korea. Currently, He s a Ph.D. canddate of Cheju Natonal Unversty, Korea. Hs major feld of study s wreless network and telematcs System. He s an Assocate Researcher of key technologes for telematcs systems at ITRC of Cheju Natonal Unversty, Korea. Gyung-Leen Park receved B.S. n Dept. of computer scence from Chung_Ang Unversty. He receved M.S. and Ph.D. from computer scence and engneerng department at the Unversty of Texas at Arlngton, respectvely. Hs research nterests nclude schedulng n parallel and dstrbuted systems, moble computng, and telematcs. In 1997, he was an assstant professor at the Unversty of Texas at Arlngton. In 1998, he joned the Dept. of computer scence and statstcs at Cheju Natonal Unversty, Korea, where he s currently an assocate professor. Currently, he s the drector of the ITRC of Cheju Natonal Unversty, Korea. Hanl Km receved the B.S, M.S, and Ph.D. degrees at Dept. of computer engneerng, Seoul Natonal Unversty, Korea. Hs research nterests nclude personalzed servce, agent system, and computer educaton. In 1995, he joned the Dept. of computer educaton at Cheju Natonal Unversty, Korea where he s currently an assocate professor. He s currently a vstng scholar at Colorado State Unversty snce 2005.