AS MOBILE wireless devices became the essential parts

Transcription

1 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 55, NO. 3, MAY A Cross-Layer Multhop Data Delvery Protocol Wth Farness Guarantees for Vehcular Networks Gökhan Korkmaz, Student Member, IEEE, Eylem Ekc, Member, IEEE, and Füsun Özgüner, Member, IEEE Abstract In ths paper, a new cross-layer communcaton protocol for vehcular Internet access along hghways s ntroduced. The objectve of the new Controlled Vehcular Internet Access (CVIA) protocol s to ncrease the end-to-end throughput whle achevng farness n bandwdth usage between road segments. To acheve ths goal, the CVIA protocol elmnates contenton n relayng packets over long dstances. CVIA creates sngle-hop vehcle clusters and mtgates the hdden node problem by dvdng the road nto segments and controllng the actve tme of each segment. Usng an analytcal throughput estmaton model, the protocol parameters are fne-tuned to provde farness among road segments. Smulaton results confrm that the proposed CVIA protocol provdes hgher throughput and better farness n multhop data delvery n vehcular networks when compared wth purely IEEE based protocols. Index Terms Computer networks, nternet, protocols, road vehcle, wreless LAN. I. INTRODUCTION AS MOBILE wreless devces became the essental parts of our lves, anytme, anywhere connectvty gans a growng mportance. Inasmuch as an average user spends hours n the traffc everyday, Internet access from vehcles s n great demand. Proposals and prototypes for vehcles supportng Internet access exst n the lterature, such as The Network Vehcle [1] and Web on Wheels [2]. In addton to these smart vehcles, the feasblty of vehcle to Internet connecton s also nvestgated n [3]. The FleetNet project [4] nvestgates the ntegraton of the Internet and vehcular networks. Ths ntegraton requres moblty support, effcent communcaton, dscovery of servces, and support of legacy applcatons. The proposed archtecture contans statonary Internet gateways (IGWs) along the road wth two nterfaces connectng vehcular networks to the Internet [5]. Vehcles communcate wth dstant IGWs va multhoppng. Ths archtecture s useful not only to connect vehcles to other networks but also to connect solated vehcle groups to each other [6]. FleetNet uses an IPv6-based addressng soluton to address the vehcles. Two approaches to solve the servce dscovery problem have been proposed n [7] and [8]. However, Manuscrpt receved August 1, 2005; revsed November 23, 2005 and December 2, Ths work was supported by OKI Electrc Ind. Co. Ltd. The revew of ths paper was coordnated by Prof. Y. Fang. The authors are wth the Department of Electrcal and Computer Engneerng, The Oho State Unversty, Columbus, OH USA (e-mal: korkmazg@ece.osu.edu; ekc@ece.osu.edu; ozguner@ece.osu.edu). Dgtal Object Identfer /TVT n these proposals, there are no specfc solutons to move data over multple hops. End-to-end throughput s one of the key parameters for vehcular Internet access systems employng an nfrastructure along the road. Although Dedcated Short Range Communcaton (DSRC) systems use the IEEE protocol as ther medum access control (MAC) layer, multhoppng wth the IEEE protocol suffers from several problems, leadng to low throughput and starvaton of packets orgnatng from vehcles far away from gateways. In ths paper, we ntroduce a new cross-layer protocol for vehcular Internet access along hghways called Controlled Vehcular Internet Access (CVIA) protocol. The proposed protocol dvdes the tme nto slots and the servce area of the gateway nto segments. The CVIA protocol controls tme slots the vehcles are allowed to transmt n, how the vehcles access the channel, and to whch vehcles the packets are sent. The CVIA protocol functons span MAC and network layers. The objectve of the new protocol s to ncrease the end-to-end throughput whle achevng farness n bandwdth usage between road segments. To acheve ths objectve, the CVIA protocol elmnates contenton n relayng packets over long dstances by formng sngle-hop clusters on-the-fly. Once vehcles send ther packets usng contenton-based methods n sngle-hop clusters, packets are relayed to ther destnatons wthout contenton. II. CVIA PROTOCOL A. Prelmnares In ths paper, we consder a vehcular network that accesses the Internet through fxed IGWs along the road. Although the wreless nterface of these gateways has a lmted wreless coverage, ther range can be ncreased wth multhop communcaton. As a result, a gateway can communcate wth a vehcle at a dstance several tmes longer than ts physcal transmsson range. The range of a gateway where t provdes Internet access servce s called the vrtual transmsson radus (VTR). We assume that gateways send perodc servce announcements to ndcate the avalablty of the servce n ther servce area. We also assume that the uplnk and the downlnk packets are transmtted over two frequency-separated channels. Vehcles are assumed to be equpped wth Global Postonng System (GPS) devces used for tme synchronzaton and obtanng vehcle postons. Vehcle postons obtaned va GPS are exchanged among one-hop neghbors. When a vehcle enters the VTR of a gateway, t regsters tself wth the gateway /$ IEEE

2 866 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 55, NO. 3, MAY 2006 Fg. 1. Slots and segments. We assume that moblty management s handled usng one of the proposed solutons n [9]. B. Defntons 1) Communcaton range R: The physcal transmsson range of the vehcles as well as the gateway. 2) VTR: The radus of the servce area of the gateway where t provdes Internet access servce. 3) Segment (S ): Fxed secton of VTR of length R. The segment closest to the gateway s denoted by S 1. 4) VTR length N: The number of segments n VTR of a gateway. 5) Tme slot j (TS j ): Tme duraton of length T slot. 6) Rato X: The rato of the mnmum length of the local packet gatherng (LPG) phase to the slot duraton. 7) Neghborng segment S + : The neghborng segment n the drecton of packet dssemnaton. S + s the neghborng segment of S closer to the gateway n the uplnk channel and the neghborng segment of S farther away from the gateway n the downlnk channel. 8) Neghborng segment S : The neghborng segment n the opposte drecton of the packet dssemnaton. 9) Interference parameter (r): r = nterference range/ R ) Actve segment: Segments where vehcle communcaton s allowed to occur n a tme slot. S s actve n TS j f ( mod r) equals to (j mod r). Note that there are r 1 nactve segments between two actve segments accordng to the actve segment defnton. In Fg. 1, when the current tme slot s T 5 and r =2, segments S 1, S 3, S 5, and S 7 become actve. In ths example, when segment (S ) s actve, ts two neghborng segments (S,S + ) become nactve. In the next tme slot (TS j+1 ), all segments change state where nactve segments become actve and actve segments become nactve. 11) Outbound temporary router (TR out ): In actve segments, the packets are gathered n vehcles closest to the segment border n the drecton of packet dssemnaton. The vehcle where the relayed packets are collected s called outbound temporary router. 12) Inbound temporary router (TR n ): At the end of an actve tme slot, TR out moves the packets to S.InS,thevehcle that receves all packets from S s called nbound temporary router. 13) Packet tran: In Secton (Request To Send/Clear To Send (RTS/CTS) usage wth fragmentaton) of the IEEE standard [10], a method s ntroduced to send several packets wth only one RTS/CTS handshake. We wll refer to ths transmsson as packet tran. In a packet tran, after the frst packet accesses the channel wth RTS/CTS handshake, the followng DATA packets are sent wthout RTS/CTS handshake. C. Overvew Our proposed CVIA protocol ams to avod the problems of the IEEE protocol n multhoppng along a hghway. It employs two vehcles as temporary routers n each segment : 1) TR out and 2) TR n. All packets enterng a segment go through TR n, and all packets leavng the segment go through TR out. We choose TR n as the closest vehcle to S and TR out as the closest vehcle to S +. The CVIA protocol uses three types of packet movements as shown n Fg. 2: 1) TR n delvers the packet tran orgnatng from other segments to TR out 2) local packets of the segments S are gathered by TR out ; and 3) TR out creates a new tran usng packets receved n (1) and ; (2) and sends t to TR n + n S +. The goals of the CVIA protocol are mtgatng the hdden node problem whle gatherng local packets, avodng contenton, and provdng farness among segments by controllng the contents of packet trans.

3 KORKMAZ et al.: MULTIHOP DATA DELIVERY PROTOCOL PROVIDING FAIRNESS FOR VEHICULAR NETWORKS 867 Fg. 2. Packet movement states. Fg. 3. State dagram of a segment. D. Packet Movement In ths secton, the detals of the packet movement n the CVIA protocol s presented wth the help of the state transton dagram shown n Fg. 3. (I) Let S be an nactve segment n tme slot j 1(TS j 1 ) and become actve when anewslot(ts j ) starts. (II) In ths new slot, vehcles n actve slots broadcast ther current postons to ther neghbors n the frst t u amount of tme. (III) In the followng phase before startng data transmsson, next temporary routers TR out,next and TR n,next are announced by TR n. (IV) After the temporary delvers the data tran orgnatng from other segments to TR out n the ntrasegment packet tran movement phase. (V) The LPG phase starts after the end of the ntrasegment packet tran. (VI) When the total number of packets queued up n TR out reaches (C/N)(N +1)or router announcement, TR n the tme left n the actve slot s just enough to move all packets n the queues, the ntersegment packet tran movement phase starts. C s the maxmum number of packets that can be delvered from S 1 to the gateway n an actve slot, and t wll be computed n Secton IV. At the begnnng of ths phase, TR out accesses the channel, creates a new ntersegment packet tran wth equal number of packets from each segment, and sends ths tran out of S to TR n +. 1) Inactve Phase: Vehcles n nactve segments do not access the channel. Inactve and actve segments are determned usng the slot number and segment number j,.e., S s actve when ( mod r) =(j mod r). Vehcles compute and j as follows: = d/r and j = t/t slot, where d s the dstance of the vehcle to the gateway and t s the tme passed snce an absolute reference pont, e.g., 12:00 P.M. Note that vehcles obtan ther postons and synchronze ther clocks usng GPS. The vehcles learn the postons of gateways from a dgtal road map database or the servce announcement packets broadcast perodcally by gateways.

4 868 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 55, NO. 3, MAY 2006 In the CVIA protocol, nactve segments are used between actve segments to ensure that the communcaton n actve segments does not nterfere wth each other. Inasmuch as the nterference range can be larger than the transmsson range, the number of nactve segments between actve segments depends on the nterference range to transmsson range rato. For example, f ths rato s 1, r s chosen to be 2, or f ths rato s 2, r s chosen to be 3. Note that when ths rato s hgher than 1, t s a very serous problem for the collson avodance (CA)-based protocols lke IEEE n multhoppng because the CTS packets cannot reach all nodes that can nterfere wth the communcaton. As a result, as the load ncreases, the number of collsons wll ncrease, leadng to a very low throughput n IEEE ) Vehcle Poston Update Phase: In the CVIA protocol, a tme nterval of t u s reserved for poston update packets (PUPs) at the begnnng of each slot. When ths phase starts, the vehcles pck a random watng tme (RWT) from [0,t u t PUP ), where t PUP s the duraton of a PUP. After an RWT, vehcles access the channel usng the dstrbuted coordnaton functon (DCF) method of the IEEE protocol. Note that because the length of a PUP s short, RTS/CTS handshake s not used before sendng PUP. 3) Temporary Router Selecton Phase: New TR out and TR n are selected by TR n after the poston update phase. In ths secton, router lfetme and safe area concepts are ntroduced frst, followed by the selecton algorthm. a) Router lfetme: Snce the topology of the vehcular network changes fast, new temporary routers must be selected perodcally. The selected routers are called TR out,next and TR n,next untl they become actve. TR out,next and TR n,next must stay nsde the segment for an amount of tme called the router lfetme. The router lfetme s dfferent for TR n and TR out. When TR out,next s selected at the begnnng of TS j, t becomes actve mmedately and stays actve untl the end of TS j. Durng ts actve tme n TS j,tr out s responsble for 1) recevng packet tran relayed by TR n, 2) gatherng local packets, and 3) creatng a new packet tran and sendng ths tran out of S to TR n +. Therefore, the router lfetme of TR out s T slot. Unlke TR out,next, TR n,next must be actve throughout TS j+1 and TS j+r.tr n s responsble for 1) recevng packet tran comng from S n TS j+1, 2) selectng and announcng TR out,next and TR n,next at the begnnng of TS j+r, and 3) relayng the packet tran to TR out n TS j+r. As a result, the router lfetme of TR n s (r +1) T slot. b) Safe area: TR out,next and TR n,next must be away from the segment border for a certan amount of dstance (x margn ) to stay nsde the segment durng the router lfetme. Snce the routers have dfferent router lfetmes, two x margn values are defned as follows: x margn+ : measured from the segment borders and assocated wth TR out ; x margn : measured from the segment borders and assocated wth TR n. The porton of the segment away from the borders by x margn+ s called the safe area of TR out, and the porton of the segment away from the borders by x margn s called the safe area of TR n. The elapsed tme snce the last PUP causes an uncertanty for the current poston of the vehcle. Therefore, x margn and safe area must be calculated separately for each vehcle v as follows: x margn,v =( t pu,v +(r +1) T slot ) V max (1) x margn+,v =( t pu,v +1 T slot ) V max (2) where t pu,v s the elapsed tme snce the last poston update from vehcle v and V max s the maxmum speed of a vehcle. If the vehcle v s dstance to segment borders s more than x margn+,v, the vehcle can safely be selected as TR out,next. Smlarly, f the vehcle v s dstance to segment borders s more than x margn,v, the vehcle can safely be selected as TR n TS j r c) Selecton of TR out,next and TR n n TS j.tr n,next: TR n s responsble for announcng TR out,next s the closest vehcle to S, and TR out closest vehcle to S + n the safe area. TR n and TR n,next as follows:,next. selected n and TR n,next s the selects TR out,next TR n computes the safe area of each vehcle n ts neghbor lst. Among the vehcles nsde the safe area, TR n selects the vehcle closest to S + as TR out,next and the vehcle closest to S as TR n,next. 4) Packet Movement Phases n Uplnk Channel: Fg. 2 depcts the packet movement phases n the uplnk channel. Intrasegment packet tran movement phase: After the vehcle poston update phase, TR n tran comng from S to TR out starts delverng the packet. To avod contenton, TR n has the hghest access prorty and wats only short nterframe space (SIFS) duraton before accessng the channel. LPG phase: After the ntersegment packet tran movement ends, the channel becomes dle for dstrbuted nterframe space (DIFS) duraton, and vehcles access the channel usng a contenton-based channel access scheme. In ths LPG phase, vehcles employ the DCF method of the IEEE standard. Each node has the same prorty, and the nodes drectly send ther packets to TR out. To decrease the number of collsons, each vehcle starts ths packet gatherng phase wth a random backoff counter. Intersegment packet tran movement phase: TR out has the hghest channel access prorty among the vehcles n the LPG phase. When the total number of packets queued up n TR out reaches (C/N)(N +1) or the tme left n the actve slot s just enough to move out all packets n the queues, TR out accesses the channel and ends the LPG phase. In the new phase, TR out forms a packet tran as descrbed n Secton III. Ths packet tran s moved to TR n + before the end of TS j. Snce there s only one TR n + n S +,TR out does not need to specfy the node ID of the destnaton n the data tran.

5 KORKMAZ et al.: MULTIHOP DATA DELIVERY PROTOCOL PROVIDING FAIRNESS FOR VEHICULAR NETWORKS 869 E. Packet Movement Phases n Downlnk Channel The downlnk packets do not need a gatherng phase because they start ther movement already at one source,.e., gateways. In each segment, some packets leave the tran, and a shorter tran keeps movng away from the gateway. As S +,TR out, and TR n are all defned accordng to the drecton of the packet dssemnaton, all namng conventons stll hold for the downlnk channel. Local packet dstrbuton and ntrasegment packet tran movement phase: TR n (or gateway) starts sendng packets to the fnal destnatons n S usng uncast transmsson. After the uncast packets are sent, TR n delvers the packet tran comng from S to TR out.tr n has the hghest access prorty and wats SIFS duraton before accessng the channel. When t d amount of tme passes from the begnnng of the tme slot, TR n stops accessng the channel. Intersegment packet tran movement phase: TR out moves the packet tran to TR n + at the end of the tme slot. III. FAIRNESS Packets orgnatng at vehcles outsde the communcaton range of the gateway must undergo several channel contentons to reach the gateway. Hence, packets that must travel shorter dstances to the gateway have advantages over such packets. To provde farness, CVIA protocol controls the contents of the packet trans leavng each segment and ntroduces farness among segments by equatng the number of packets from each segment n packet trans. Unlke per-flow farness, segmentbased farness s scalable and does not requre mantenance of any state nformaton beyond a tme slot, whch are mportant features n a network wth a fast changng topology. In the context of ths paper, farness s defned as provdng equal throughput to all segments. Varous quanttatve measures are proposed for farness n the lterature. We choose Jan s farness ndex (FI) [11] due to ts populaton sze ndependence, boundedness (0 FI 1), contnuty, and metrc ndependence propertes. FI s defned as FI = ( N =1 Thr ) 2 /N N =1 Thr2, where Thr s the throughput of segment. To equate the number of packets from each segment, temporary routers utlze a far queung scheme whle formng packet trans. A separate queue s mantaned for each segment s packets n outbound temporary routers. When an outbound temporary router receves a packet, t nserts the packet to the queue dedcated to ts segment. Whle formng a packet tran, packets are chosen usng the round robn method where one packet s taken from each nonempty queue n each round. IV. CHANNEL CAPACITY C AND FAIRNESS AS A FUNCTION OF THE LENGTH OF THE GATHERING PHASE The mnmum length of the LPG phase s crtcal to provde farness among segments. We ntroduce the mnmum LPG phase length parameter, whch forces an actve segment to be n the LPG phase at least X fracton of the actve tme slot. In ths analyss, we am to calculate the optmum length of the mnmum LPG length parameter (X opt ), whch results n the same number of packets from each segment so that FI =1. Our analyss s performed for the worst case condton when the total demand (TD) s above channel capacty C and the demand of each segment s above C/N packets per actve tme slot. Note that channel capacty s defned for each actve tme slot so that when channel capacty s C packets, t means that, at most, C packets can be delvered to the base staton n each actve tme slot. C s computed usng (13). Once X opt s computed for the worst case condton, t s used as the mnmum LPG phase length parameter for all loads and all segments. These are the results of usng X opt and far queung mechansm. 1) When the demand of each segment s above C/N, the FI becomes 1,.e., CVIA provdes throughput guarantee of C/N to each segment ndependent of the demands of other segments. 2) To satsfy (1), each S s able to move out (C/N)(N +1) packets n each actve tme slot. Ths amount corresponds to the packets comng from outer segments and the local packets gathered from S. If the throughput demands of some outer segments (S +1 to S N )arelower than C/N, S can fll the excess capacty usng local packets. When ths s the case, the length of the LPG phase becomes longer than the mnmum LPG phase length parameter (X opt ), and t s clear that FI wll not be 1. However, ths stuaton arses as a result of demand dstrbuton of segments, and t does not mean that CVIA protocol s unfar. Each segment has a maxmum output lmt of ((C/N)(N + 1)) packets. The temporary routers queue up a maxmum number of packets, all of whch can be sent out n one actve slot: TR out queues up (C/N)(N +1) packets, and TR n queues up (C/N)(N ) packets. Therefore, they do not have any packets n ther queues at the end of ther lfetme, and no handoff s necessary when a temporary router leaves the segment. A. Calculaton of X opt n S 1 The bottleneck for the total end-to-end throughput s the segment closest to the gateway where all packet traffc passes through. Therefore, the saturaton throughput of S 1 determnes the channel capacty C of the system. The number of packets receved by the gateway from segment S 1 can be dvded nto two dstnct parts. Num outer,1 : The packets contaned n the packet tran orgnatng from outer segments (S 2 to S N ). Num gather,1 : The local packets orgnatng from S 1 and receved n the LPG phase. To provde farness, two condtons must be satsfed at saturaton pont. 1) The number of packets orgnatng from each outer segment (Num ) n the packet tran should be equal so that Num =(Num outer,1 /N 1). 2) The length of the LPG phase n S 1 should be long enough so that Num gather,1 =(Num outer /N 1).

6 870 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 55, NO. 3, MAY 2006 Fg. 4. Actve slot and packet tran n S 1. Recall that packets from the outer segments reach the frst segment n a packet tran, and each temporary router along the way employs far queung. As a result, the number of packets orgnatng from each outer segment (Num ) wll be equal n the packet tran f the length of the LPG phase s long enough n each segment to gather suffcent number of packets such that Num gather, (Num outer,1 /N 1). As the same mnmum gatherng phase length s employed n all segments, f we can satsfy condton 2, condton 1 s also satsfed. In ths secton, Num outer,1 and Num gather,1 are computed as afunctonofx. Then, the optmum X value (X opt ) satsfyng Num gather,1 (X opt )= Num outer,1(x opt ) (3) N 1 s calculated. Fnally, the total throughput and the FI are computed as a functon of X. B. Calculaton of Num outer,1 As shown n Fg. 4, (1 X) fracton of the actve slot tme (T slot ) s allocated to the packets orgnatng from the outer segments and X fracton to the LPG phase. The packets from outer segments use 0.5(1 X) of the actve slot tme to enter to S 1 and the remanng 0.5(1 X) fracton to reach the gateway. Whle these packets move n a packet tran, they access the channel wthout any contenton. Furthermore, RTS/CTS handshake overhead apples only to the begnnng of the packet tran. The total number of packets orgnatng from the outer segments (S 2 to S N ) and receved by the gateway n an actve slot s computed as 0.5(1 X) Tslot T to Num outer,1 (X) = (4) T tp where T to s equal to T DIFS + T RTS + T SIFS + T CTS and T tp s equal to T SIFS + T DATA + T SIFS + T ACK. T RTS, T CTS, T DATA, and T ACK are the lengths of the RTS, CTS, DATA, and ACK packets, respectvely; T DIFS and T SIFS are the dstrbuted and short nterframe spaces, respectvely. C. Calculaton of Num gather,1 In the LPG phase, vehcles access the channel usng the IEEE protocol. Snce the neghborng segments are nactve durng ths phase, we assume that hdden termnals do not exst. To compute the saturaton throughput of the IEEE protocol, Banch presents a Markov model where backoff wndow of each staton s modeled by a bdmensonal process [12]. As a result of Banch s analyss, the probablty of transmttng a packet n a generc backoff slot tme (τ) and the probablty that a transmtted packet undergoes collson (p) are gven as 2(1 2p) τ = (1 2p)(W +1)+pW (1 (2p) m ) (5) p =1 (1 τ) n 1 (6) where W s the mnmum contenton wndow sze, n s the number of vehcles, and m s the maxmum number of retransmssons. Once ths system of nonlnear equatons s solved numercally, the probablty P tr that there s at least one transmsson n a backoff slot tme and the probablty P s that a transmsson s successful gven that at least one staton s transmttng can be obtaned as P tr =1 (1 τ) n and P s = nτ(1 τ) n 1 /P tr, respectvely. We defne the normalzed system throughput S as the fracton of channel tme used for successful packet transmssons. where S = E tsp E t (7) E tsp = P tr P s T p (8) E t =(1 P tr )σ + P tr P s T p + P tr (1 P s )T c. (9) In (7), E tsp s the expected tme to transmt one successful packet, and E t s the expected tme between two successful packet transmssons. In (8) and (9), σ s the backoff slot duraton, T p s the length of a successful packet transmsson, and T c s the length of the tme wasted n case of a collson. T c

7 KORKMAZ et al.: MULTIHOP DATA DELIVERY PROTOCOL PROVIDING FAIRNESS FOR VEHICULAR NETWORKS 871 Fg. 5. FI Scenaro I. and T p can be computed as follows: T p = T RTS +3 T SIFS + T CTS + T DATA + T ACK + T DIFS, and T c = T RTS + T DIFS. Once S s computed, the number of packets orgnatng from the frst segment and receved by the gateway can be calculated as Num gather,1 (X) = S X T slot. (10) T p D. Determnng the Optmum Length of the Gatherng Phase (X opt ) Snce Num outer,1 and Num gather,1 are expressed as functons of X, optmum X value can be calculated usng (3). T p (0.5 T slot T to ) X opt =. (11) (N 1) S T tp T slot +0.5 T p T slot The number of packets from the outer segments can be calculated as Num =(Num outer,1 /N 1). Consequently, the FI can be represented as a functon of X. ForX X opt FI(X) = (Num gather,1(x)+num outer,1 (X)) 2 ( ). (12) N Num 2 outer,1(x) gather,1 (X)+Num2 N 1 Note that FI(X opt )=1. The number of packets receved by the gateway n an actve slot s computed by addng the number of packets from the outer segments and the packets from the frst segment,.e., N total (X) =Num outer,1 (X)+Num gather,1 (X). (13) Note that channel capacty C of an actve slot s C = N total (X opt ). V. P ERFORMANCE EVALUATION To evaluate the performance of our CVIA protocol and the IEEE protocol, we have developed a Wreless Smulator (WS) usng the event-drven smulaton lbrary Clent- Sde Image Map (CSIM) [13]. In all smulaton scenaros, smulated vehcles move on a lnear hghway segment wth two lanes, one for each drecton of traffc flow. The vehcles randomly enter the servce area wth exponentally dstrbuted nterarrval tmes. On the average, there are 34 vehcles/km per lane. Each vehcle s assgned a speed from a Gaussan dstrbuton wth a mean of 90 km/h and a standard devaton of 5 km/h at the begnnng of the smulaton. The assgned speeds do not change durng smulaton. The common parameters of the smulatons are as follows: transmsson range = 350 m, data rate =27Mbps, payload = 2312 or 500 bytes, base protocol = a, smulaton tme = 10 s, smulaton repettons =20, maxmum number of packet retrals =10, and nterference range to transmsson range rato =1. The parameters of the CVIA protocol are as follows: T slot = 100 ms, N =4or 8, and r =2. Other parameters of the MAC layer and the physcal layer are taken from the ASTM E standard document [14]. A. Scenaros To examne the effects of packet length and VTR on the performance of the network, three dfferent scenaros are smulated for varous packet loads. The parameters for these scenaros are as follows: Scenaro I: payload = 2304 bytes, VTR =4R (N = 4); Scenaro II: payload = 500 bytes, VTR =4R (N =4); and Scenaro III: payload = 2304 bytes, VTR =8R (N =8). B. Results 1) Optmalty of X opt : Ths secton evaluates the effect of the mnmum LPG phase length on FI at saturaton. Fg. 5 ndcates that theoretcal analyss s successful n estmatng X opt =0.18 because the peak value of the smulaton curve matches our X opt estmaton calculated usng (11). When X< X opt, the outer segments have larger shares n the total throughput. On the other hand, when X>X opt, S 1 domnates the total throughput and causes starvaton n outer segments. In addton to Scenaro I, (11) s used to estmate the X opt values of Scenaros II and III as and 0.085, respectvely. Our smulatons show that these estmates also match the smulaton results. 2) Uplnk Channel: In ths secton, we examne the performance of the protocols n the uplnk channel. X opt value gven

8 872 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 55, NO. 3, MAY 2006 Fg. 6. Throughput Scenaro I. Fg. 8. Detaled segment throughputs Scenaro I. Fg. 7. Throughput Scenaro II. Fg. 9. Detaled segment throughputs Scenaro II. n Secton V-B1 s employed as the mnmum length of the gatherng phase n each scenaro. Due to space lmtaton, we present the results of Scenaro III n common fgures only. a) Throughput and farness: Fgs. 6 and 7 depct the endto-end throughput of IEEE and CVIA protocols n Scenaros I and II as the nput load ncreases. In these fgures, deal throughput curve s drawn n addton to the protocol performance curves wth a maxmum throughput value (MTV), where MTV =1/(T DIFS + T DATA + T SIFS + T ACK ). The throughput curves of both protocols follow the deal throughput curve when the packet load s low. When the load s ncreased, the CVIA protocol reaches a saturaton pont. Ths saturaton pont s the channel usage capacty of the protocol and successfully estmated by the theoretcal analyss usng (13). The saturaton throughput of CVIA s slghtly larger than half of the maxmum deal throughput value. The packets away from the transmsson range of the gateway hold the channel frst to enter to the transmsson range and then to reach the gateway. As a result, f all packets orgnate from outsde the transmsson range of the gateway, MTV/2 becomes the effectve approxmate maxmum throughput value. As some packets orgnate from the frst segment, the resultng throughput becomes slghtly larger than half of the maxmum deal throughput. Although the CVIA protocol reaches a saturaton pont, the throughput of the IEEE protocol keeps ncreasng. To better understand the reason behnd ths ncrease, we need to refer to Fgs. 8 and 9. These fgures show the throughput of S 1 and the average throughput of all the outer segments (S 2 to S N ). In IEEE , the outer segments experence starvaton, whereas the end-to-end throughput becomes greatly domnated by packets of the frst segment. On the other hand, the CVIA protocol dstrbutes the throughput to all segments farly, and ths farness s not affected by the offered load. As a result, although the total end-to-end uplnk throughput of the IEEE protocol ncreases wth ncreasng packet load, t suffers from a very serous problem: Increased load causes starvaton for the outer segments. The effects of ths phenomenon on farness are depcted n Fg. 10, where FI s very close to 1 for all loads n the CVIA protocol. However, FI of the IEEE protocol drops drastcally when the load s ncreased.

9 KORKMAZ et al.: MULTIHOP DATA DELIVERY PROTOCOL PROVIDING FAIRNESS FOR VEHICULAR NETWORKS 873 Fg. 10. Farness. Fg. 12. Delay Scenaro I. Fg. 11. TPI. Fg. 13. Delay Scenaro II. Fg. 11 shows the throughput percentage ncrease (TPI) n all three scenaros. TPI corresponds to the amount of throughput gan obtaned by employng our proposed CVIA protocol nstead of the IEEE protocol. It s computed as PI = (T CVIA T /T ) 100, where T CVIA and T are the throughputs of the CVIA and IEEE protocols, respectvely. The vertcal lnes on the graph mark the nput loads where the FI of the IEEE protocol drops to 0.5. Note that FI of CVIA s close to 1 when FI of IEEE s 0.5 as shown n Fg. 10. To the rght of those lnes, the outermost segments are effectvely dsconnected from the gateway n the IEEE protocol. The percentage ncrease starts to decrease especally to the rght sde of those lnes because the packets from the frst segment domnate the throughput n the IEEE protocol. When the total throughput s domnated by the packets from the frst segment, the total throughput ncreases. The reason for ths ncrease s that when a packet orgnates from the transmsson range of the gateway (S 1 ), t holds the channel only once. However, f t orgnates from outsde of the transmsson range, t holds the channel twce, frst to enter to the transmsson range and second to reach to the gateway. b) Delay: In the context of ths paper, delay refers to the tme elapsed between the nstant the packet enters the transmsson queue of the source and the recepton tme of the packet by the fnal destnaton. As shown n Fgs. 12 and 13, the average packet delay of the protocols starts to ncrease when the protocols reach ther saturaton load. Due to space lmtaton, the results of Scenaro III are not presented. After the saturaton pont, the queue lengths start to ncrease, leadng to hgh delay values. The starvaton n outer segments also affects the delay performance. To better understand the delay performance of the uplnk channel, we also plot the delay curve of the packets only from S 1 and the average delay curve of the packets orgnatng from the outer segments. Note that although all three delay curves of the CVIA protocol are close to each other after the saturaton pont, the delay curves of the IEEE protocol show dfferent behavors. In the IEEE

10 874 IEEE TRANSACTIONS ON VEHICULAR TECHNOLOGY, VOL. 55, NO. 3, MAY 2006 through smulatons that the proposed protocol has hgher endto-end throughput when compared wth the IEEE protocol. Moreover, the throughput gan obtaned usng the proposed CVIA protocol s hghest for short payload scenaro because the CVIA protocol avods relatvely large RTS/CTS overhead. Fnally, our protocol dstrbutes the throughput farly among segments, although the vehcles far away from the gateway suffer from starvaton under the IEEE protocol. In our future work, the CVIA protocol wll be modfed to provde throughput and delay guarantees for ndvdual vehcles. Fg. 14. Faled packets protocol, as the throughput becomes domnated by the frst hop packets, the overall delay curve comes closer to the delay curve of the packets from S 1. However, the delay of the outer segments keeps ncreasng as we ncrease the load. As a result, n the IEEE protocol, the average delay of packets orgnatng from the outer segments s the hghest n all three scenaros. c) Packet falure rate: Fg. 14 shows the number of packets dropped n one second due to packet collsons. When the load s ncreased, the IEEE protocol starts droppng packets because more nodes become actve per unt tme, and ths creates more hdden nodes for each transmsson. On the other hand, because the CVIA protocol avods hdden nodes whle gatherng packets and then moves these packets n packet trans wthout contenton, the probablty of a packet collson s much lower than that of the IEEE protocol. In the CVIA protocol, the contenton-based channel access s used n LPG phase. Accordng to (6), the probablty of packet collson s approxmately 0.3 even for the worst case stuaton. To drop a packet, ten retrals must be attempted. As a result, the packet droppng probablty of the CVIA protocol s estmated as = ) Downlnk Channel: Smulaton results of the downlnk channel show that employng CVIA protocol ncreases the endto-end throughput. Due to space lmtaton, downlnk results are not presented n ths paper. REFERENCES [1] R. Lnd, R. Schumacher, R. Reger, R. Olney, H. Yen, M. Laur, and R. Freeman, The network vehcle A glmpse nto the future of moble multmeda, IEEE Aerosp. Electron. Syst. Mag.,vol.14,no.9,pp.27 32, Sep [2] A. Jameel, M. Stuempfle, D. Jang, and A. Fuchs, Web on wheels: Toward Internet-enabled cars, Computer, vol. 31, no. 1, pp , Jan [3] T. Osafune, K. Monden, S. Fukuzawa, and S. Matsu, Performance measurement of moble ad hoc network for applcaton to Internet ITS (ntellgent transportaton system), n Proc. SAINT,Jan.2004,pp [4] FleetNet homepage. (2006, Apr. 23). [Onlne]. Avalable: tu-harburg.de/fleetnet/ndex.html [5] M. Bechler, W. Franz, and L. Wolf, Moble Internet access n FleetNet, n 13. Fachtagung KVS, [Onlne]. Avalable: [6] J.-J. Tchouto, K. Kutzner, B. Bochow, T. Luckenbach, M. Bechler, and L. Wolf, Connectng vehcle scatternets by Internet-connected gateways, n Proc. Multrado Multmeda Commun., [Onlne]. Avalable: [7] M. Bechler, O. Storz, W. Franz, and L. Wolf, Effcent dscovery of Internet gateways n future vehcular communcaton systems, n Proc. 57th IEEE Semannu. VTC Sprng, 2003, pp [8] P. Ratanchandan and R. Kravets. (2003). A hybrd approach to Internet connectvty for moble ad hoc networks, n Proc. IEEE WCNC, pp [Onlne]. Avalable: cteseer.st.psu.edu/ ratanchandan03hybrd.html [9] E. Perera, V. Svaraman, and A. Senevratne, Survey on network moblty support, Moble Comput. Commun. Rev., vol. 8, no. 2, pp. 7 19, [10] Wreless LAN Medum Access Control (MAC) and Physcal Layer (PHY) Specfcatons. (1999). [Onlne]. Avalable: (a.k.a. ISO/IEC :1999(E)). ANSI/IEEE Std [11] R. Jan, W. Hawe, and D. Chu, A quanttatve measure of farness and dscrmnaton for resource allocaton n shared computer systems, Dgtal Equpment Corp., Maynard, MA, DEC-TR-3001, Sep [12] G. Banch, Performance analyss of the IEEE dstrbuted coordnaton functon, IEEE J. Sel. Areas Commun., vol.18, no.3,pp , Mar [13] H. Schwetman, CSIM User s Gude Rev. 1, Mcroelectron. Comput. Technol. Corp., Austn, TX, Tech. Rep. ACT , Rev. 1. [14] ASTM E Standard Specfcaton for Telecommuncatons and Informaton Exchange Between Roadsde and Vehcle Systems 5 GHz Band Dedcated Short Range Communcatons (DSRC) Medum Access Control (MAC) and Physcal Layer (PHY) Specfcatons VI. CONCLUSION In ths paper, a new far channel access and routng strategy for vehcular Internet access along hghways s ntroduced. The CVIA protocol mtgates the hdden node problem, avods contenton and unnecessary RTS/CTS overhead whle movng packets among road segments, and provdes farness among segments by controllng the contents of the packet trans. To provde farness, an analytcal model s developed to estmate the throughput of each segment. By usng ths analytcal model, an optmum length for the LPG phase s computed to equate the number of packets delvered from each segment. It s shown Gökhan Korkmaz (S 05) receved the B.S. degree n electrcal engneerng from Mddle East Techncal Unversty, Ankara, Turkey, n 2000 and the M.S. degree n electrcal engneerng from The Oho State Unversty, Columbus, n He s currently workng toward the Ph.D. degree n the Department of Electrcal and Computer Engneerng, Oho State Unversty. He s currently a Research Assocate n the Department of Electrcal and Computer Engneerng, Oho State Unversty. Hs man research nterest s wreless networks, wth a specal emphass on vehcular communcaton systems. He s partcularly nterested n broadcastng, ntersecton collson warnng systems, and Internet access n vehcular networks.

11 KORKMAZ et al.: MULTIHOP DATA DELIVERY PROTOCOL PROVIDING FAIRNESS FOR VEHICULAR NETWORKS 875 Eylem Ekc (S 99 M 02) receved the B.S. and M.S. degrees n computer engneerng from Bogazc Unversty, Istanbul, Turkey, n 1997 and 1998, respectvely, and the Ph.D. degree n electrcal and computer engneerng from Georga Insttute of Technology, Atlanta, n Currently, he s an Assstant Professor n the Department of Electrcal and Computer Engneerng, Oho State Unversty, Columbus. Hs current research nterests nclude wreless sensor networks, next generaton wreless systems, and vehcular communcaton systems, wth a focus on routng and medum access control protocols, resource management, and analyss of network archtectures and protocols. He also conducts research on nterfacng of dssmlar networks. Füsun Özgüner (S 74 M 75) receved the M.S. degree n electrcal engneerng from the Istanbul Techncal Unversty, Istanbul, Turkey, n 1972 and the Ph.D. degree n electrcal engneerng from the Unversty of Illnos, Urbana-Champagn, n She was wth IBM T. J. Watson Research Center, Yorktown Heghts, NY, wth the Desgn Automaton group for one year and joned the faculty at the Department of Electrcal Engneerng, Istanbul Techncal Unversty, n Snce January 1981, she has been wth The Oho State Unversty, Columbus, where she s presently a Professor and the Interm Char of Electrcal and Computer Engneerng. Her current research nterests are parallel and faulttolerant archtectures, heterogeneous dstrbuted computng, reconfguraton and communcaton n parallel archtectures, real-tme parallel computng and communcaton, and wreless networks. Dr. Özgüner has served as an Assocate Edtor of the IEEE TRANSACTIONS ON COMPUTERS and on program commttees of several nternatonal conferences.