A Probabilistic Routing Protocol in VANET

Transcription

1 MoMM 29 Full Papers Proceedngs of MoMM29 A Probablstc Routng Protocol n VANET Gongun Yan Coputer Scence Departent Old Donon Unversty Norfolk, VA ygongun@cs.odu.edu Stephan Olaru Coputer Scence Departent Old Donon Unversty Norfolk, VA olaru@cs.odu.edu Shaharuddn Salleh Unverst Teknolog Malaysa Johor Bahru, Malaysa ss@el.ut.y ABSTRACT The key attrbute that dstngushes Vehcular Ad hoc Networks (VANET) fro Moble Ad hoc Networks (MANET) s scale. To wt, whle MANET nvolve up to one hundred nodes and are short lved, beng deployed n support of specal-purpose operatons, VANET networks nvolve llons of vehcles on thousands of kloeters of hghways and cty streets. Also, beng sson-drven, MANET oblty s nherently lted by the applcaton at hand. Indeed, n a search-and-rescue sson a search party s deployed to look, say, for chldren lost n the woods. It stands to reason that n such a context the oblty of ndvdual nodes s not arbtrary, but rather restrcted by the search strategy pleented. In ost MANET applcatons, such as the one ust entoned, oblty occurs at low speed. By contrast, VANET networks nvolve vehcles that ove at hgh speed, often well beyond what s reasonable or legally stpulated. Gven the scale of ts oblty and nuber of actors nvolved, the topology of VANET s changng constantly and, as esult, both ndvdual lnks and routng paths are nherently unstable. Motvated by ths latter trus, our frst an contrbuton s to propose a probablty odel for lnk duraton based on realstc vehcular dynacs and rado propagaton assuptons. We then go on to show how the proposed odel can be ncorporated n outng protocol that results n paths that are easer to construct and antan, both locally and globally. Extensve sulaton results confr that our probablstc routng protocol results n paths are that easer to antan. Keywords VANET, routng, probablty, path antenance. INTRODUCTION The last few years have wtnessed the rapd developent of ntellgent Transportaton Systes (ITS) and Vehcular Work supported, n part, by NSF grant CNS Persson to ake dgtal or hard copes of all or part of ths work for personal or classroo use s granted wthout fee provded that copes are not ade or dstrbuted for proft or coercal advantage and that copes bear ths notce and the full ctaton on the frst page. To copy otherwse, to republsh, to post on servers or to redstrbute to lsts, requres pror specfc persson and/or a fee. MoMM29, Deceber 4-6, 29, Kuala Lupur, Malaysa Copyrght 29 ACM /9/2...$.. Ad hoc NETworks (VANET) that prose to revolutonze the way we drve. In the not-so-dstant future vehcles equpped wth coputng, councaton and sensng capabltes wll be organzed nto a ubqutous and pervasve vehcular network that can provde nuerous servces to travelers, rangng fro proved drvng safety and cofort, to locaton-specfc servces, to delverng ulteda content on deand, and to other slar value-added servce. Indeed, the fact of beng networked together prootes car-to-car councatons, even between cars that are tens of les apart. The short transsson range andated by DSRC (Dedcated Short Range Councatons) has as a natural corollary the fact that councatons between vehcles are usually ulthop. For exaple, consder a car that travels down an nterstate and whose passengers are nterested n vewng a partcular ove. The varous blocks of ths ove happen to be avalable at varous other cars on the nterstate, often les away. Cars n varous lanes ove at dfferent speed, akng the underlyng network hghly dynac. In such a network, ndvdual councaton lnks are short-lved and the routng paths that rely on a ulttude of such lnks are nherently unstable. The splest routng strategy s blnd floodng based. However, floodng-based routng s known to waste network resources, snce each packet s transtted ultple tes, often unnecessarly. The broadcast stor that s resulted by floodng ethod has been shown to draatcally reduce network throughput. In an effort to reduce the effects of floodng, the past years have seen a flurry of well-deserved actvty n the area of desgnng lght-weght routng protocols n both MANET and VANET. A word of cauton s n order here: due to ther fundaental dfferences already entoned, routng algorth devsed specfcally for MANET cannot be sply ported to VANET. In support of routng over large dstances, n addton to vehcle to vehcle councatons VANET reles (or wll rely n the near future) on pre-deployed roadsde nfrastructure that facltate not only access to the Internet but, ndeed, can also assst wth the task of propagatng packets between vehcles, even n sparse traffc. The roadsde nfrastructure relays or even buffers packets untl a sutable vehcle s avalable to carry t further. In addton to roadsde nfrastructure, the geographc poston of ndvdual vehcles s used to optze the routng process. Such a feature s enabled by the phenoenal prolferaton of coercally-avalable GPS recevers that cost a few tens of dollars. Vehcles aware of ther own geographc poston as well of that of ther neghbors can select a greedy/effcent routng path to forward 79

2 Proceedngs of MoMM29 MoMM 29 Full Papers packets. However, n spte of ther ntrnsc appeal, nether the nfrastructure-based nor the geographc poston based routng ethods are perfect and there s a lot of roo for proveent. One coponent that sees to be lackng s an understandng of the volatlty of ndvdual lnks n a routng paths. Wth ths n nd, the otvaton of ths paper s to provde elable routng schee that takes nto account the lfete of ndvdual lnks. Snce oblty s one of the reasons that VANET dffers so sgnfcantly fro MANET, we analyze the relatonshp between key oblty paraeters (sgned velocty and acceleraton) to buld a novel probablty odel that can predct the lfete (duraton) ndvdual councaton lnks. Furtherore, we propose a probablty odel based on the oblty analyss. We beleve that an understandng of the probablty dstrbuton of lnk duraton s fundaental to descrbng the dynacs of vehcles councatons. Wth such an understandng frly n hand, one can select routng paths that are lkely to be stable and are easer to antan at both the local and global level. 2. RELATED WORK Based on the attrbutes (probablty, oblty, poston, neghbors), we can classfy the exstng VANET routng protocols nto four categores: probablty, oblty, poston, and floodng based routng. There are soe nfrastructure based routng, but we are nterested n the nonnfrastructured network,.e. the adhoc network. Probablty Based Routng: Jang et al. [] present a routng ethod, called REAR, based on the recepton probablty of alar essages. The selecton of the next hop s based on the recepton probablty. The recepton probablty s coputed usng the relatonshp between the packet loss rate and the receved sgnal strength. Yang et al. [2] develop outng protocol called Connectvty Aware Routng (CAR). The connectvty odel s on the probablty coputaton on oad whch s parttoned nto grds/cells. Then the probablty to copute the connecton between two nodes s to copute the probablty that the dstance between the two nodes s wthn a certan value (transsson range of wreless). Yan et al. [3, 4] propose a tcket based routng strategy n VANET by usng the expected duraton of lnks. The expected lnk duraton s coputed by a probablty odel. A Tcket-Based Probng wth Stablty constraned(tbp-ss) routng ethod s also proposed n ([4]). The key paraeter used to select outng s based on the ean duraton of a lnk (defned as stablty) whch s coputed by the probablty odel. Usng a dvde and conquer approach, each optal routng lnk s selected and results an optzed routng path whch s coposed by each optzed lnk. Vehcles are equpped wth a DSRC/IEEE 82-based wreless transcever (e.g. IEEE 82.p) and a GPS devce. Moblty Based Routng: Moblty s one of the aor attrbutes that sets VANET apart fro other networks. [5] presents outng ethod by groupng vehcles accordng to velocty. Vehcles are grouped nto four dfferent groups based on ther velocty vectors (speed wth drectons). [6] presents a enhanced routng protocol to adapt the hgh oblty of vehcles, three oblty paraeters: poston, drecton and speed. [7] addresses outng algorth based on the nterdependences of average vehcle speed, traffc densty and the traffc qualty (n ters of congeston). Geographc Poston Based Routng: Poston based routng protocols have receved a great deal of attenton n VANET. CarNet ([8]) uses grd to propagate packets. The grd s defned on the bass of geographc poston of nodes. CarNet, as announced by authors, can support any applcatons: IP connectvty, cooperatve hghway congeston ontorng, fleet trackng, and dscovery of nearby ponts of nterest. [9] proposes a ethod for constructng network groups accordng to lane poston and evaluate the proposed ethod by sulaton as well. [] presents a zone floodng algorth and a zone routng algorth. A zone s defned as a geographc area, for exaple a 5 eter secton of oad. Vehcles that are n the zone are allowed to broadcast packets. Otherwse the receved packets wll be dropped. [] presents a greedy routng strategy based on geographc postons of vehcles. [2] descrbes outng protocol, called ROVER (RObust VEhcular Routng) based on Zones whch are defned as postons of vehcles whch are equpped wth GPS recevers. Slarly, [3] proposes eactve algorth whch uses zone-lke group routng ethod. The control traffc packets are selectvely transtted by the selected nodes, called gateways. All the ebers n the zone can read and process the packet, but do not retranst the packets. Only gateway nodes retranst packets between zones,.e. gateway to gateway councaton. Floodng Based Routng: In floodng routng protocol, the basc dea s to (flood) broadcast packets through the whole network. [4] broadcasts RREQ to fnd a routng path to gateway and a RREP s sent to acknowledge the routng path. [6] uses RREQ to explore routng path as well. These types of routng were grafted on AODV ([5]) whch s ntally proposed for MANET. [6] proposes outng ethod whch extends the orgnal floodng by acknowledgng floodng essages. [7] aggressvely propagates the packets to the nodes near the axu transsson range by only allowng these nodes to rebroadcast the packets. 3. THE PROBABILITY MODEL 3. Coputng Lnk Duraton The duraton of a lnk s the te nterval durng whch an establshed councaton lnk between two vehcles contnues to exst. The an goal of ths secton s to derve analytcal expressons of the (te) duraton of a lnk and expressons for the probablty dstrbuton of the (te) duraton of a lnk. Assue, wthout loss of generalty, that at te t =, a councaton lnk s establshed between two vehcles and ovng n the sae drecton or n the opposte drecton, wth ahead of. Let the rando varable X denote the dstance separatng the two vehcles at lnk setup te. Mndful of the axu DSRC transsson range 8

3 MoMM 29 Full Papers Proceedngs of MoMM29 constrant, we have X < 3. () The dstance between two consecutve cars passng the sae pont and travelng n the sae drecton s defned as headway dstance. The headway dstance plays a fundaental role n understandng traffc flow and n ensurng travel safety and related ssues. The log-noral dstrbuton was proposed to odel headways under the car-followng odel [8]. It s nterestng to enton that our results agree wth those of a slar experent conducted and reported ndependently by Puan [9]. For ore detaled dscusson, refer our prevous research [2]. It s portant to note that, n general, X s the convoluton of ndependent headway dstances, and t s well known that the convoluton of ndependent log-noral rando varables can be approxated by a log-noral rando varable [2]. For defnteness, we assue that X has paraeters µ and σ. We assue that the speed lt on the roadway s v and that no vehcle wll run faster than v. For t, we defne a(t), the acceleraton of the vehcle at te t as follows: f a() =, then a(t) = for all t ; f a() >, then a(t) = f a() <, then a(t) = a() for t v v() a() otherwse a() for t v() a() otherwse In other words, (2) and (3) ndcate that as long as the vehcle has not reached the axu speed v or has not stopped (n case a() < ), ts acceleraton reans a(). However, once the vehcle reaches the speed lt ts speed (or has stopped), ts acceleraton becoes. Gven a generc vehcle wth ntal speed v(), the nstantaneous speed v(t) at te t s defned as v(t) = v() + (2) (3) a(u)du, (4) where for all u [, t], a(u) s the nstantaneous acceleraton at te u defned by above. Now, (2) and (3) and (4), cobned ply that f a() =, then v(t) = v() for all t ; f a() >, then v() + a()t for t v v() v(t) = a() v otherwse f a() <, then v() + a()t for t v() v(t) = a() otherwse Slarly, the dstance that our generc vehcle travels n the te nterval [, t] s defned as S(t) = (5) (6) v(x)dx, (7) where v(x) was defned above. We now return to our vehcles and. To splfy the notaton, we wrte v = v (), a = a () and v = v (), a = a (). The nstantaneous speeds and acceleratons v (t) and a (t), respectvely, v (t) and a (t) are obtaned by sutably nstantatng (2), (3), (5), (6). Now, (7) guarantees that the dstances traversed n the te nterval [, t] by vehcle and are, respectvely, and S (t) = S (t) = v (x)dx (8) v (x)dx. (9) Assung that at connecton setup (.e. te ) the dstance between the two vehcles was x, t follows that the dstance between and at te t can be wrtten as S (t) S (t) + X. () It s portant to notce that () defnes a sgned dstance: ndeed, f at te t, S (t) S (t) + X >, then vehcle s ahead of ; otherwse, vehcle s ahead of. For later reference, we fnd t convenent to defne the ndcator functon I(, ) ntended to capture nforaton about whch of the two vehcles s ahead when the councaton lnk between the breaks, f S (t) S I(, ) = (t) + X > () otherwse Gven that DSRC lnks break at 3 eters, t follows that when the lnk breaks the followng relaton holds: S (t) S (t) + X = 3 I(, ). (2) We refer the reader to Fgure where the varous possble cobnatons of v, v, a, a are llustrated, assung that vehcles and ove n the sae drecton,.e. v, v >. We now defne a nuber of te nstances that wll be used n our analyss: provded that v v a a >, we wrte defne t β as follows v a v t β = a n v a undefned t α = v v a a ; (3) f v a > and v a < f frac v a > and v a <, v a } f v a > and v otherwse. a > (4) slarly, defne t γ as follows ax v t γ = a, v a } f v a > and v a > undefned otherwse. (5) 8

4 Proceedngs of MoMM29 MoMM 29 Full Papers defne t ε as follows v v a v v t ε = a n v v a undefned, v v a } f v v a f v a > and v v a < f v a > and v v a < > and v v a > otherwse (6) defne t ζ as follows ax v v t ζ = a, v v a } f v v a > and v v a > undefned otherwse (7) It s portant to note that t α t β t γ} and t ε, t ζ } only depend on the speeds and acceleraton of the two vehcles at connecton setup te and on the value of the speed lt v. Therefore, vehcles and can use the followng algorth to predct the te (f any) when the establshed lnk wll break: Algorth 3.: FndLnkBreakTe(v, v, a, a ) procedure ProbeRoutngPath(te) f S (te) S (te) + X 3 then t = +(, te] else t = (, te] return (t) an regon = [, + ] x use v, v, a, a to copute all the tes t α, t β, t γ, t ε, t ζ } for each t x do regon = regon BreakTeRegon(t) deterne the value of I(x, y) based on the value of regon solve the roots of the equaton S (t) S (t) + X = I(, ) 3 te deterne the rght root output (te) Dscusson: The algorth FndLnkBreakTe takes the four sgned oblty paraeters: v, v, a, a as nput and returns the lnk duraton te. Frst, t α, t β, t γ, t ε, t ζ are coputed as descrbed by (3) (7) above. For each of the, we check the te regon where the lnk ay break. We ark the regon where the lnk wll break as a postve regon and the regon that the lnk wll not break as a negatve regon. Then, we ntersect these postve and negatve regons to fnd the sallest regon where the lnk wll break. If t turns out that the connecton breaks, the correspondng value of the ndcator functon I(, ) s coputed. Fnally, we can solve the equaton to fnd the lnk duraton te. Due to the page lt we lst the exact lnk duraton soluton t for cases a n appendx A. A detaled dervaton of the all cases can be found n [2]. 3.2 Coputng the Probablty of Lnk Duraton The an goal of ths secton s to derve the probablty dstrbuton functon of lnk duraton. Startng fro the general for of dstance functon, we can derve the probablty functon of lnk duraton by usng the dstrbuton functon of the rando varable X. If any of v (t) and v (t) s not a constant, the dstance functon wll be a quadratc polynoal. We wrte v (t) = v () + a t and, slarly, the dstance functon S (t) = v () + a xdx = v ()t + 2 at2. (8) Therefore, as already noted, S (t) S (t) s, n general, a quadratc polynoal. Suppose S (t) S (t) = 2 t 2 +bt+c where a, b, c R. Substtute S (t) S (t) n (2), to obtan the quadratc equaton 2 art2 + bt + c + X I(, ) 3 =. (9) Clearly, t exsts because the lnk breaks at te t. Therefore, the soluton ust be a square root functon of X f. We wrte the soluton as b+ b 2 2 (c+x I(,) 3 a t = r as t b b 2 2 (c+x I(,) 3 as t 2 = b + d b 2 2 (c + X I(, ) 3 (2) where d = when the lnk duraton s t ; d = 2 when the lnk duraton s t 2. If =, we can easly obtan t = I(,) 3 X c. We observe that the lnk duraton s a b rando varable T. b+d b 2 2 (c+x I(,) 3) T = (2) I(,) 3 X c a b r = (22) Let G T be the probablty dstrbuton functon of T. For every postve t, we wrte G T (t) = Pr[T t}]. (23) Snce T s obvously contnuous, now, f (23) allows us to wrte G T (t) = Pr[T t}] = Pr[ b + d b 2 2 (c + X I(, ) 3 t}] a r X ( = 2 art2 + bt + c I(, ) 3) for > X ( 2 art2 + bt + c I(, ) 3) for < ( FX ( = art2 + bt + c I(, ) 3) ) ( > F X ( a 2 rt 2 + bt + c I(, ) 3) ) < where F X s the probablty dstrbuton functon of X. If = (23) allows us to wrte G T (t) = Pr[T t}] I(, ) 3 X c = Pr[ t}] b X (bt + c I(, ) 3) for b > = X (bt + c I(, ) 3) for b < FX ( (bt + c I(, ) 3)) for b > = F X ( (bt + c I(, ) 3)) for b < Once we have the probablty functon of a lnk, we can easly copute the probablty of a lnk by sutably nstantatng t n G T (t). Specfcally, the t s usually the routng path duraton requreent EDur. 82

5 MoMM 29 Full Papers 4. OUR PROPOSED PROTOCOL In ths secton, we wll show how the proposed schee works. We assue that the syste s coposed by vehcles whch nstall the sae transcevers, GPS and wth unque denttes. For the purpose of ths paper, we assue that the traffc s oderate n ntensty, snce sparse traffc renders networkng as we descrbe t possble. We also assued that only the vehcles on the sae drecton wll be recruted to propagate packets because the opposte vehcles break the routng lnk sooner than the sae drectonal vehcles. 4. Probng the Routng Path Due to hgh oblty, the topology of the network s constantly changng. Therefore, eactve routng path probng s needed,.e. outng path searchng s on-deand. In the lterature, there are several reactve routng protocols, for exaple AODV, DSR, DSDV etc. whch were dscussed earler. These protocols fnd outng path on the bass of reachablty nstead of qualty of the routng path. We propose a new routng path search protocol on the bass of qualty of the routng path: the duraton of the routng path, as well as the reachablty. Ths s because the duraton of outng path s the ost dffcult and portant etrc to consder when routng packets n the presence of hgh vehcular oblty. In the routng process llustrated n Fgure 2(a).a, we use the unque vehcle ID as the network address. The source vehcle wll send out a probng request Prb (a control packet): PROB. Snce the PROB s broadcast by wreless channel, all vehcles n the transsson range wll receve t and copute the dstance fro the sender, the duraton of the lnk and the probablty of the lnk. The coputaton s based on the prevous dscusson. An acknowledgent packet AckP wll be constructed by usng the coputed dstance, duraton, and probablty of the lnk. The AckP s sent back to the sender who wll collect ultple AckP and select the best lnk. The best lnk s the one wth hghest probablty and havng larger duraton than the routng path duraton requreent EDur. The sender only sends a confr packet CfP to the best lnk whch wll further explore the routng path by ultcastng the Prb. When the destnaton node receves the probng packet, t wll ternate the probng and send an acknowledgent packet back to the source node along the newly fored routng path. Ths copletes the routng path exploraton stage. The route we found s optal n ters of long duraton. 4.2 Routng Path Mantenance Usually, the route antenance s the followng: when a route s about to break, two strateges can be eployed: global repar or local repar. Frst, the lnks that causes the breakage of the routng path wll be locally repared/replaced. Second, a backup route wll be constructed globally before the breakage of the current routng path and wll be swtched as the current route. The repar procedures are based on the predcton of the lnk duraton dscussed n Secton 3. and the probablty of the lnk duraton dscussed n Secton 3.2. By usng ths predcton, we can repar the current path before t s broken. In ths way, we enhance the relablty of the routng path and antan avalablty Local Repar In our protocol, data packets wll nclude a oblty header Proceedngs of MoMM29 whch ncludes: the type of the packet ( type n Fgure 2(b)), the duraton of the lnk ( dur ), the routng path duraton requreent ( EDur ), the probablty of the lnk ( Pr ), the nuber of hops ( Hop ), the source and destnaton address ( Src, Dst ), the current node s d ( ID ), the prevous and next router ( pre, next ), the recever of the lnk s sgned speed and acceleraton ( r v, r a ), the sender of the lnk s sgned speed and acceleraton ( s v, s a ), the dstance of the lnk ( ds ). Durng the councaton, each recever of a lnk wll ontor the connectvty condton of the lnk by coputng the duraton and the probablty of the lnk. If the lnk s about to break, the recever wll send a control packet RBRK (Routng BReaK). The sender wll send a Upd packet whose hop s set to. The Upd packet only explore the one-hop neghbors to fnd eplaceent lnk Global Repar All the vehcles along the routng path know the expected expraton te of the path. These vehcles do not need to check the connectvty of the routng path f the vehcles ove at constant speed. But the predcton of the routng path needs to be updated due to the dynacs of vehcular oblty where vehcles accelerate and decelerate randoly. Therefore, the connectvty of routng path needs to be updated as well. The frequency of the updatng packets t upd s deterned as follows. There s a fxed frequency: t uf, for exaple seconds. Consder that there wll be soe short connecton. Suppose p pnt s a certan percentage of the routng path duraton requreent EDur. The frequency of updatng packets s t uf f t uf p pnt EDur, otherwse, t s p pnt EDur f t uf > p pnt EDur. As shown n Fgure 2(a).b, the source node wll send a update packet Upd for every t upd seconds. When a node receves the Upd, t wll copute the dstance fro the sender, the duraton of the lnk and the probablty of the lnk. The coputaton s based on the prevous dscusson and wll not be repeated here. An acknowledgent packet AckU wll be constructed by usng the coputed dstance, duraton, and probablty of the lnk. The AckU s sent back to the sender who wll collect ultple AckU and wll select the best lnk whch s the one wth hghest probablty and havng larger duraton than the routng path duraton requreent EDur. The sender only sends a confr packet CfU to the best lnk whch wll further explore the routng path by ultcastng the Upd. When the destnaton node receves the update packet, t wll ternate the updatng process and wll send an acknowledgent packet back to the source node along the latest routng path. Ths copletes the routng path update. 5. SIMULATIONS State of the art sulators for vehcular wreless networks nvolve two coponents: a oblty sulator and a wreless network sulator. These two coponents ay be tghtly ntegrated (two coponents n one sulator) or else loosely ntegrated (two coponents n two separated sulators but connected by trace fles). In our sulator, we use looselyntegrated enttes. Our oblty sulatons were perfored under the oblty sulaton platfor of Treber and Helbng [22]. The oblty we used s IDM [23] whch s close to the realty drvng behavors. We recorded the trace fles of 83

6 Proceedngs of MoMM29 MoMM 29 Full Papers vehcle oblty and then ported the nto NS-2.3. In NS-2, each node represents one vehcle n the oblty sulatons, ovng based on the represented vehcle oveent trace. Nodes n NS-2 are also enttes of wreless councaton nodes ntegratng network protocol stacks. The general paraeters we used n sulatons are detaled n Table. The total nuber of vehcles n s varyng because n our oblty sulaton, vehcles ay enter and ext. The nuber of vehcles on the road depends on the headway dstance. Table : The envronent confgure Nae Value The nuber of vehcles (n) -2 Applcaton CBR Network protocol IEEE 82. Transsson range 3 Network connectons n/3 Sulaton ap Hghway/urban Road length 5 K Traffc densty 5 vehcles/hour Average speed 28 /s Acceleraton range [,2] /s 2 Intal acceleraton /s 2 Intal speed 25 /s Moblty odel IDM [23] Car followng odel way-pont We sulated our routng protocol and have copared t wth DSR [24] and ROMSGP [25]. DSR s a well-known ondeand routng protocol and ROMSGP s a oblty-based routng protocol. Frst we nvestgated the average duraton of the routng path. The source and destnaton nodes are randoly selected along the vehcles on the hghway. For each routng path, we recorded the path duraton and obtaned the average path duraton. Fgure 3(a) shows the average path duraton for the three protocols. As t turns out, our protocol provdes ore stable paths than the other two protocols. Ths s because we select the ost durable routng lnks. We also are nterested n the control essage overhead. Fgure 3(a) shows the average overhead of control essages for each of the selected protocols. Copared wth DSR and ROMSGP, our proposal s ore effcent. For any applcatons, throughput s one of key factor. We copared our protocol wth the other two protocols n ters of throughput. Fgure 3(a) shows the average throughput of the three protocols. The throughput decreases when the speed of vehcles ncreases. More lnks wll break when speed s hgh, thereafter, the total channels decrease and the bandwdth decreases. But, even n ths scenaro, our proposed protocol stll outperfors both DSR and ROMSGP snce fewer routng lnks wll break under our protocol. We sulated the proposed protocol and copared t wth a probablty-based routng protocol CAR [2] and a wellknown greedy-propagaton routng protocol GPSR [7]. Snce the latter two protocols are sulated on urban ap. We use SUMO [26] as our oblty sulator to record oveents of all vehcles as a trace fle and then, port the trace fle nto ns-2. To copare wth the CAR and GPSR protocol, we changed our sulaton paraeters accordng to the paraeters used n [2]. Fgure 4(a) shows the average delay of vehcles. The delay decreases when the data rate ncreases because the packets have to wat f the routng lnks are broken, untl new lnks are constructed. The proposed protocol gves lowest delay aong the three protocols because less routng lnks wll break. We nvestgated the packet delvery rato to see how any packets were successfully delvered. As shown n Fgure 4(b), both CAR and our protocol have hgh packet delvery rato and our proposal even have a hgher delvery rato because of our routng path antenance. We have also copared the throughput of the three protocols. Fgure 4(c) shows the throughput coparson aong the three protocols. Our protocol has hgher throughput than CAR. 6. CONCLUSION AND FUTURE WORK In ths paper we proposed outng protocol whch proves the relablty of councaton n VANET. The key dea s to fnd outng path wth hghest duraton. Accordng to the current state of oblty of vehcles, the duraton of a lnk can be predcted by our oblty and probablty odel. Based on ths odel, the ost stable lnks can be selected. The routng path s also updated and antaned usng the oblty and probablty odel. Copared wth the on-deand protocol DSR [24] and the oblty-based protocol ROMSGP [25], the proposed schee has been shown effectve n ters of path duraton, reduced control overhead and hgh throughput. Addtonally, copared wth the probablty-based routng protocol CAR [2] and the greedypropagaton routng protocol GPSR [7], our protocol has been shown to feature low delay, hgh packet delver rato and hgh throughput. The future work ncludes: ) extend the oblty odel to nclude opposte vehcles; 2) study the councaton under sparse traffc. 7. REFERENCES [] H. Jang, H. Guo, and L. Chen, Relable and effcent alar essage routng n vanet, n ICDCSW 8: Proceedngs of the 28 The 28th Internatonal Conference on Dstrbuted Coputng Systes Workshops. Washngton, DC, USA: IEEE Coputer Socety, 28, pp [2] Q. Yang, A. L, and P. Agrawal, Connectvty aware routng n vehcular networks, n Wreless Councatons and Networkng Conference, 28. WCNC 28. IEEE, 3 28-Aprl 3 28, pp [3] G. Yan and S. Olaru, A probablstc analyss of path stablty n vehcular ad hoc networks, Technque Report n Coputer Scence of Old Donon Unversty, 29. [4] G. Yan, S. Olaru, and S. El-Tawab, Tcket-based relable routng n vanet, n Proceedngs of the Frst Internatonal Workshop on Intellgent Vehcular Networks (InVeNET 29), Macau, Chna, Oct 29. [5] T. Taleb, E. Sakhaee, A. Jaalpour, K. Hashoto, N. Kato, and Y. Neoto, A stable routng protocol to support ts servces n vanet networks, Vehcular Technology, IEEE Transactons on, vol. 56, no. 6, pp , Nov. 27. [6] O. Abed, M. Fathy, and J. Taghloo, Enhancng aodv routng protocol usng oblty paraeters n vanet, n AICCSA 8: Proceedngs of the 28 84

7 MoMM 29 Full Papers Proceedngs of MoMM29 IEEE/ACS Internatonal Conference on Coputer Systes and Applcatons. Washngton, DC, USA: IEEE Coputer Socety, 28, pp [7] H. F. Wedde, S. Lehnhoff, and B. van Bonn, Hghly dynac and scalable vanet routng for avodng traffc congestons, n VANET 7: Proceedngs of the fourth ACM nternatonal workshop on Vehcular ad hoc networks. New York, NY, USA: ACM, 27, pp [8] R. Morrs, J. Jannott, F. Kaashoek, J. L, and D. Decouto, Carnet: A scalable ad hoc wreless network syste, n In Proceedngs of the 9th ACM SIGOPS European workshop: Beyond the PC: New Challenges for the Operatng Syste. ACM Press, 2, pp [9] T. Kato, K. Kadowak, T. Kota, and K. Sato, Routng and address assgnent usng lane/poston nforaton n a vehcular ad hoc network, n APSCC 8: Proceedngs of the 28 IEEE Asa-Pacfc Servces Coputng Conference. Washngton, DC, USA: IEEE Coputer Socety, 28, pp [] J. Bronsted and L. Krstensen, Specfcaton and perforance evaluaton of two zone dssenaton protocols for vehcular ad-hoc networks, n Sulaton Syposu, th Annual, Aprl 26, pp. 2 pp.. [] J. Gong, C.-Z. Xu, and J. Holle, Predctve drectonal greedy routng n vehcular ad hoc networks, n ICDCSW 7: Proceedngs of the 27th Internatonal Conference on Dstrbuted Coputng Systes Workshops. Washngton, DC, USA: IEEE Coputer Socety, 27, p. 2. [2] M. Khl, M. Schtu, T. Ekeroth, and M. Rozenberg, Relable Geographcal Multcast Routng n Vehcular Ad-Hoc Networks. Sprnger Berln / Hedelberg, 27. [3] S. Moen and M. Fathy, Vanet s councaton, n Spread Spectru Technques and Applcatons, 28. ISSSTA 8. IEEE th Internatonal Syposu on, Aug. 28, pp [4] V. Naboodr and L. Gao, Predcton-based routng for vehcular ad hoc networks, Vehcular Technology, IEEE Transactons on, vol. 56, no. 4, pp , July 27. [5] C. E. Perkns, E. M. Beldng-Royer, and S. R. Das, Ad hoc on-deand dstance vector (aodv) routng, RFC Experental 356, July 23. [Onlne]. Avalable: [6] S. Bswas, R. Tatchkou, and F. Don, Vehcle-to-vehcle wreless councaton protocols for enhancng hghway traffc safety, Councatons Magazne, IEEE, vol. 44, no., pp , Jan. 26. [7] B. Karp and H. T. Kung, GPSR: greedy pereter stateless routng for wreless networks, n MobCo : Proceedngs of the 6th annual nternatonal conference on Moble coputng and networkng. New York, NY, USA: ACM, 2, pp [8] I. Greenberg, The log-noral dstrbuton of headways, Australan Road Research, vol. 2, no. 7, pp. 4 8, 966. [9] O. C. Puan, M. R. Hann, A. A. Chk, and Isal, Drverś car followng headway on sngle carrageway roads, Malaysan Journal of Cvl Engneerng (MJCE), vol. 6, no. 2, Jan 24. [2] G. Yan, Ph.d dssertaton: Provng locaton secruty n vanet, Coputer Scence Departent, Old Donon Unversty, 29. [2] Z. Lu, J. Alhana, and R. McGoran, Approxatng log-noral su dstrbutons wth power log-noral dstrbuton, IEEE Transactons on Vehcular Technology, vol. 43, no. 4, pp , July 28. [22] Moblty traffc sulator, [23] M. Treber, A. Hennecke, and D. Helbng, Congested traffc states n eprcal observatons and croscopc sulatons, Physcal Revew E, vol. 62, p. 85, 2. [24] D. B. Johnson, D. A. Maltz, and Y. C. Hu, The dynac source routng protocol for oble ad hoc networks (dsr), Publshed Onlne, IETF MANET Workng Group, Tech. Rep., February 27. [25] T. Taleb, E. Sakhaee, A. Jaalpour, K. Hashoto, N. Kato, and Y. Neoto, A stable routng protocol to support ts servces n vanet networks, Vehcular Technology, IEEE Transactons on, vol. 56, no. 6, pp , Nov. 27. [26] Open source, Sulaton of urban oblty, APPENDIX A. LINK DURATION CASES Due to the page lt, we lst lnk duraton case a. For ore detaled dscusson or the rest of cases, refer our prevous research [2]. In Case a, only t β and t ε exst, as shown n fgure.a. We can obtan the lnk duraton te t. Let a > > a ; v > v > } be condton one, arked as c ; a > > a ; v > v > } be condton two, arked as c 2. v r + vr 2+2(3 X), c ; t t α} v r vr 2 2(3+X), c ; t α < t t ε} t = (v v ) (v v ) 2 2a (3+X+ v v t 2 ε ) 3+x+ v v 2 t ε + v 2 t β a, c ; t ε < t < t β } v, c ; t β < t} v r vr 2 2(3+X), c 2; t t α} v r + vr 2+2(3 X), c 2 ; t α < t t ε } (v v ) (v v ) 2 2a (3 X+ v v t 2 ε ) 3 X+ v v t 2 ε v 2 t β a, c 2; t ε < t t β } v, c 2 ; t β < t} 85

8 Proceedngs of MoMM29 MoMM 29 Full Papers (a) β β (b) β (c) (d) β (e) (f) (g) β γ α β γ (h) Fgure : Illustratng the sae drecton case. Sender Recever Prb AckP CfP (a) Sender Recever Upd AckU CfU (b) type dur EDur Pr hop Src Dst ID pre next r_v r_a s_v s_a ds Others (len, chksu, prorty, sg, etc) (a) protocol (b) The packet header Fgure 2: The protocol and packet header Average path duraton (s) DSR ROMSGP Control overhead (# of control packets) [^6] 5 DSR ROMSGP Average path duraton (s) DSR ROMSGP The speed (/s) (a) Duraton The speed (/s) (b) Control essage overhead The speed (/s) (c) Throughput Fgure 3: Coparson wth oblty-based routng 25 5 Delay (sec) 2 5 CAR GPRS Delvery rato CAR GPRS Throughput (kb/s) CAR GPRS Data sendng rate (kb/s) (a) Delay Data sendng rate (kb/s) (b) Delvery rato Data sendng rate (kb/s) (c) Throughput. Fgure 4: Coparson wth probablty-based routng 86