MORA: a Movement-Based Routing Algorithm for Vehicle Ad Hoc Networks

Transcription

1 : a Movement-Base Routing Algorithm for Vehicle A Hoc Networks Fabrizio Granelli, Senior Member, Giulia Boato, Member, an Dzmitry Kliazovich, Stuent Member Abstract Recent interest in car-to-car communications an networking has lea to the efinition of the concept of Vehicle A-hoc NETwork (VANET) as an infrastructure-free a-hoc networking solution in the automotive scenario. The requirement for proviing reliable an efficient routing schemes in presence of relative movement motivates the proposal of, a movement-base routing algorithm for VANETs. The algorithm is completely istribute, since noes nee to communicate only with irect neighbors within their transmission range, an it exploits a specific metric, which exploits not only the position, but also the irection of movement of vehicles. Extensive simulations evaluating the propose protocol an results of comparison with state-of-the-art methos emonstrate that provies a promising an robust basis for esigning a routing strategy suitable for the automotive scenario. Inex Terms Vehicle Networking, Routing, Movement-Base Routing, A-Hoc Networks. I. INTRODUCTION ngoing research in the fiel of ICT for automotive O applications is riving the integration of electronic an communication evices proviing several ae-value services, such as positioning an navigation, automatic toll payments, monitoring of the status of the vehicle, etc. A growing interest within this scenario lies in the possibility of enabling vehicles to access the Internet or other network commoities, or more in general communicating an collaborating. This feature is envisage to be implemente into two ways: (i) by the eployment of proper communication infrastructure along the roas to act as gateways to the Internet, or (ii) by the implementation of the so-calle Vehicle A-hoc NETwork (VANET). In the last scenario, routing support shoul be provie by each vehicle belonging to the VANET without the nee for specific communication infrastructure. Currently, the Car--Car Communication Consortium [] ientifie guielines for proviing vehicle-to-vehicle communications as well as a reference protocol architecture, but i not efine channel an traffic moels, channel usage, an routing algorithms yet. This leaves the floor to further stuy an proposals, especially in the context of routing. This work is partially supporte by EU in the framework of the SAFESPOT Integrate Project. F Granelli, G. Boato, an D. Kliazovich are with the Dept. of Information an Communication Technology (DIT), University of Trento, Via Sommarive, I-85 Trento, [granelli, boato, klezovic]@it.unitn.it. Inee, the basic concept of VANETs erives from the well-known moel of the mobile a-hoc networks (MANETs), infrastructure-less networks where wireless hosts communicate with each other in the absence of a fixe infrastructure. Multihop ata communication in VANETs is usually provie via location-base a hoc routing protocols [], a class of multihop routing for a hoc networks. Traitionally, multi-hop routing for MANETs can be classifie into proactive an reactive algorithms: in proactive routing algorithms, each noe in the mobile a hoc network maintains a routing table that contains the paths to all possible estinations. If the network topology locally changes, all routing tables throughout the network have to be upate. If the noes in the network are reasonably mobile, the overhea of control messages to upate the routing tables becomes prohibitive. Reactive routing algorithms, on the other han, fin routes only on eman. Routes are esigne when they are neee, in orer to minimize the communication overhea. A etaile review of routing algorithms in mobile a hoc networks can be foun in []-[], which were lately integrate by many contributions (see for instance [5]-[6]). In this framework, an interesting approach is represente by position-base routing algorithms, which require information about the physical position of the participating noes an it is exactly the class of algorithms envisage to be implemente in VANETs, ue to the continuous localization process performe by GPS evices equippe on vehicles. In such schemes (like Distance Routing Effect Algorithm for Mobility (DREAM) [7] an Location Aie Routing (LAR) [8]), the forwaring ecision is primarily base on the position of the packet estination an the position of the noe's immeiate one-hop neighbors. A etaile survey of protocols that o use geographic location in the routing ecision is presente in [9]- []. One of the most promising routing approaches for VANETs is location base greey forwaring routing, the example of which is presente in []. In Greey Perimeter Stateless Routing (GPSR) protocol all packets transmitte onto the network are marke by the originator with their estination s locations. As the result, a forwaring noe can make a locally optimal routing ecision. Specifically, the noe assume to know the exact positions of its neighbors forwars the packet to the neighbor closest to the estination. As the result, the main avantage of greey forwaring is in its reliance on knowlege of the immeiate neighbors of the forwaring noe. When a packet reaches the region when greey

2 forwaring is not possible ue to specifics of the topology, GPSR recovers traversing the topology graph aroun the perimeter of this region. However, if only position information is use, it may be possible to lose some goo caniates to forwar the packet. Moreover, the knowlege of noe s position coul not be sufficient in a network with frequent topological changes, such as a VANET. In such a situation it is important to guarantee high stability of the links an therefore robustness of the routing protocol. To provie a solution to the above-mentione problems, we propose an alternative movement-base routing algorithm (), which exploits not only the position, but also the irection of motion of mobile noes. Awareness of a noe s movement irection implemente in routing represents an attempt to fin a solution to this critical problem an to match the requirements of inter-vehicle communication. Current state of the art alreay inclues some proposals for routing in VANETs (see for instance [5]-[7]). However the problem remains uner a hot iscussion rising a number of open issues neee to be solve. The structure of the paper is the following: Section II escribes the metho, while extensive simulations are reporte in Section III. Finally, Section IV conclues the paper. II. THE PROPOSED ALGORITHM A. Preliminaries In a position-base routing algorithm, each noe makes a ecision to which neighbor to forwar the message base only on the location of itself, its neighboring noes, an the intene estination. Therefore, the system can be ecentralize, more robust an easier to set up an operate. In our approach, consiering the specific automotive scenario, this ecision is taken consiering also which irection neighbors are moving in. Since VANETs are subject to frequent topology changes, the life time of connections between hosts varies appreciably. Our goal is to exploit information about moving irections of the forwaring noes in orer to route the ata over a path resulting from locally optimal routing ecision. In literature, there are a lot of ifferent strategies a noe can employ to properly select a neighbor for forwaring a given packet [9]. However, none of them takes into consieration that noes in a hoc network are moving in irections that can introuce unpreictable changes in the network topology, thus affecting alreay establishe routes an network connectivity in general. In the efinition of the routing algorithm, we assume that each vehicle is moving along a regular route, i.e. its movement pattern remains constant uring packet transmission. Moreover, we neglect the impact of errors in the techniques use for position estimation of the vehicles leaving it as an open issue for future investigation. The basic concept of this work, in the framework of a generic MANET, was presente in []. The metric use in (Movement-Base Routing Algorithm) is a linear combination of the number of hops an a target functional, which can be inepenently calculate by each noe, as escribe in the following paragraphs. The functional F. The core iea of the approach is to evelop a functional which epens on the istance of forwaring car from the line connecting the source an the estination,, an on the noe s movement irection. This functional is require to be implemente in a istribute way allowing any vehicle to calculate it. The target functional shoul reach its absolute maxima in the case the noe is moving on an it shoul ecrease as the istance from increases. Moreover, the more a noe moves towars, the higher shoul be its value, i.e. for a fixe istance from the functional shoul have a maximum if the noe is moving perpenicularly to. Let be a reference istance metric, chosen on the basis of the application context. Let x = be the aimensional istance of the current noe from an y = l the aimensional istance from the estination of the intersection point between an its perpenicular starting from the noe s current position (see Fig. ). The functional F is a function of x [, ] an α [ π, π ], where α represents an angle between the line of the movement irection an the perpenicular line to (see Fig. ). Source noe i α noe j Fig.. Graphical representation of parameters use by. In orer to ensure the targete properties, we choose the functional F as follows: ( x δ ) γ α x α F = sin e + cos e () where δ an γ are two parameters set on the basis of the application, which simply vary the curvature of F, ajusting the weight associate with noe s movement irection, δ efines the value of x corresponing to the relative maximum along the x axis an γ leas to a smoother or steeper behavior own to zero. Such a efinition of F assures more weight to noes moving on, an also to noes moving towars (see Fig. ) as require above. In fact for x = there are absolute maximums, for α = ±π respectively; α l l Destination

3 for < x < ε (ε arbitrarily small) the tren is the same as above; for x the function ecreases; for x = δ there is a relative maximum corresponing to α = ; for x δ a, δ + b ] ( a an b constants.5.5 pi [ efine with the choice of δ an γ ) there is a maximum corresponing to α =. pi/ alpha pi/ pi Fig.. Plot of the functional F. The functional F can be sample an put into a look up table. In this way, each noe oes not nee to calculate F at any iteration, but it can easily obtain the value corresponing to a given combination of x an α with a simple an fast table lookup. The metric m. Another egree of freeom of the metric employe in is the weight assigne to each noe, which can be use to represent traffic conitions, application constraints, etc. The goal of the weighting function is to obtain a fair istribution of the available resources through the overall network. For the purpose of the paper, the function W, efine for y [, y s ] where y s is the y-component of the source noe, is given by: w( x, <. W = () log w( x,. w( x, where w [,] is the weight of noe i with coorinates x, y. Now the following metric can be efine, for y [, y s ]: m y, α ) = ( W + Fδ, λ α )) () where both W an F [, ], an therefore m y, [, ]. Due to the fact that x an y are the coorinates of noe i, an α epens on the noe i, in following sections we refer to m y, an m i without istinction. The reaer shoul note that, by choosing x such metric, the higher the value of m i the higher the probability noe i is inclue into the active route from source to estination. The presente way of noe weighting provies a possibility to inclue other than location an movement parameters into routing. The weight wxy (, ) associate with the noe can be calculate base on such parameters like a level of noe s congestion, an outgoing ate rate, available power resources, etc. For example, in case noe i is congeste an therefore w, then W. B. The routing protocol In position-base routing algorithms, usually short probe messages are sent into the network in orer to etermine the position of the estination noe, which is use for route establishment. In more etails, the sener floos a route establishment request into the network or its part. The estination replies to the sener with a route reply packet incluing such information like its location. After a route reply has reache the sener, the ata payloa can be transmitte using position-base routing algorithms. routing uses flooing for estination iscovery like most of existing routing protocols. The sener inclues its location information into the route request flooe into the network. Upon the reception of a route request from the sener, the estination noe generates a route reply message which is route using metric m efine in Eq.(). On every hop, the current noe receiving it polls for information its neighboring noes, consiering only those with the higher values of y in orer to avoi loops (y is relate to the istance from the estination as in Section II, A). The coorinates of the source noe, coorinates of the estination noe, position of the noe last forware the packet, as well as its moving irection, are inclue into every protocol message. As a result, each noe is able to obtain metric m for itself as well for its immeiate neighbors. The values for an α use in functional F calculation presente in Eq.() are obtaine as follows: yi m xi q =, α = cos () ± + m ) ist ( where m an q are calculate using coorinates of the source an estination noes ( xs, ys ) an ( x, y ), respectively: y ys m =, (5) x x q y y s y x s = s (6) s x xs where ist is the istance of the noe from along the irection of movement. The probe message is then forware to the neighbor with the highest value of m (see Section II.A), attaching path information.

4 Since all vehicles will probably use GPS technology in the near future, we try to exploit the availability of up-to-ate information about positions an moving irections of the source noe, estination noe as well as noes locate along the line an their immeiate neighbors. The frequency of the upates is epenant on the particular implementation of the routing protocol. In this paper we consier two possible implementations of routing: -Stanalone. This implementation, referre to as, separates the framework of the propose routing protocol into a stanalone routing layer. As a result, location an movement information is carrie by only routing protocol messages (such as Route Request an Route Repl. The main rawback of the stanalone implementation is that position information is not upate uring ata packet exchange. -Link layer integrate. In orer to overcome the upate limitations of the stanalone approach, integration of protocol with the MAC protocol at the link layer is consiere as a moification referre to as +. In aition to the features of the stanalone implementation, + inclues the location an movement information into the orinary MAC protocol heaers, which carry signaling or ata payloa. This technique enables a ynamic upate of such information along the entire ata path for every transmitte packet, thus avoiing waste of available communication resources. III. PERFORMANCE EVALUATION Performance evaluation of the propose routing protocol was performe by simulations using GloMoSim. [8] network simulator. GloMoSim is a scalable simulation environment for wireless mobile networks base on the Parsec parallel iscrete-event simulation library. GloMoSim is chosen out of the set of available network simulators to the fact of the availability of physical layer moels fairly approximating real-worl behavior as well as for extensive support of mobility in a hoc networks (in terms of movement patterns an routing algorithms). IEEE 8. physical layer stanar is chosen for the set of conucte experiments. An aitional software moule enabling functionality was inserte into a stanar Glomosim package. In orer to achieve integration between routing plane an link layer protocol require by +, the corresponing moifications were performe for the MAC protocol. The propagation of route request is implemente using flooing moel. However, after coorinates of a estination are iscovere, the route reply message as well as ata payloa packets are route using techniques. In case a noe can not fin the route to estination (which is probably cause by wrong/change coorinates of the estination), it sens a route error message to source. A. General simulation scenario Simulations are performe for five routing protocols:,, LAR,, an +. The results are obtaine for variable number of noes, their moving spee as well as transmission range. The noes are uniformly place onto a two-imensional terrain of x square meters. The number of network noes is chosen to be in orer to achieve a satisfactory connectivity. Simulations use transmission range values equal to,,, an 5 meters. As a result, ata communication between any pair of noes can occupy from to 7 hops. The sener an the estination noes are chosen ranomly. Stanar FTP client operating over TCP protocol was chosen as a traffic source application. For evaluation of routing overhea, the FTP client was configure to prouce bulks of packets in large (.5 secon) intervals of time. After each bulk transmission, the routing table as well as the table with neighbor noes is cleare for all the noes. This requires initiation of route iscovery for every generate bulk of packets. In other scenarios, FTP application performs uninterrupte ata transfer for up to the en of simulation which lasts for secons. The ranom waypoint mobility moel (with pause time equal to zero) is use: each noe performs several moves uring the simulation time without remaining static between moves. The noes move with an average spee of 5 an 5 meters-per-secon or 5 an 9 km/h. Our simulation results are average over runs with ifferent sees of the ranom generator. The results where the communication between ranomly chosen sener an receiver noes was not possible ue to isconnecte topology (which happene rarel are exclue. B. Routing Overhea In this section, routing overhea is compare against other available routing techniques. The overhea is efine as the number of routing packets (requests, replies, route failures) sent over the entire network within a single burst transmission. Forwaring of routing control packet is consiere as a separate transmission. Figure unerlines that implementation behaves similar to flooing routing algorithms (for average spees of 5 an 5 meters, respectivel. This is motivate by the fact that estination iscovery is performe using route request flooing into the network. A slight enhancement over the flooing curve comes from the ifference in the route reply propagation, which is route using noe movement information. Routing overhea becomes consierably lower when LAR is use limiting the region of the network the route request is flooe. However, we recall the fact that the protocol oes not limit the technique use for estination iscovery to flooing. Inee, operation starts from the point when the position of the source noe an the estination noes are available, which happens when the route request message reaches the estination noe. It allows an implementation of any existing route request propagation scheme, thus leaing to the corresponing avantages. C. Performance vs Range The throughput performance versus transmission range for ifferent mobility levels is illustrate in Fig.. FTP source always achieves lowest throughput in case routing is use. fixes the routes for route reply propagation as well as for subsequent ata communication on the en-to-en

5 5 basis. As a result, any changes in connectivity between any neighboring noes create a route failure, which can only be resolve by generation of a new route iscovery initiate at the source noe. protocol emonstrates better throughput performance if compare with. A per-hop base routing appears to be more stable than the one fixe on the en-to-en routes. The routes etermine by protocol are more stable in presence of mobility. However, the fact that coorinates of estination are etermine only uring the route request phase limits the performance of in case the estination moves relatively far from its initial position (etermine uring the route iscover. This problem is solve in + version of the protocol, which is an example of close integration between routing plane an the MAC protocol layer. The location of the estination as well as intermeiate noes is ynamically upate with every ata or control packet transmission. As a result, + is almost insensitive to mobility in the presence of continuous ata exchange along the route. The ifference in performance of evaluate protocols is better shown for low values of transmission range, while for high transmission ranges communication between noes can be achieve through a lower number of hops, thus limiting performance to similar throughput values. Figure 5 presents the performance of evaluate routing protocol versus noe s mobility. appears to be the most sensitive to mobility. The performance of + is consistently stable for low as well as for high noes moving spees. D. Highway Scenario The results presente above show goo performance of the propose protocol in a generic a-hoc network scenario. However, in orer to ensure its performance gain in VANETs, we performe evaluation in a highway simulation scenario presente in Fig. 6. It consists of three cars R, R an R moving on the right lane an n cars on the left lane moving in the opposite irection. The parameter n is chosen to be large enough to ensure there is always a left-lane car within the transmission range of R, R, an R. The istance between every pair of cars riving the same lane is chosen to be 5 meters, while the transmission range is fixe to 7 meters. All the simulate cars are moving with a preefine spee ranging from to m/s (7 km/h). The uration of each experiment is limite by min uring which the cars riving at 7 km/h cover the istance of. km. Left lane Right lane L R 5 m 5 m L L Ln R 5 m 5 m Fig. 6. Simulate highway scenario. In the first set of experiments, the R car running an FTP source communicates with the R car. This implies a two-hop communication performe between cars moving in the same R irection. The obtaine performance results (Fig. 7) show that all evaluate protocols achieve similar level of the throughput. This is mainly an outcome of a high stability of the route R R R where the communication is performe. FTP Throughput (bps).5 x Spee (m/s) Fig. 7. FTP transfer between cars moving in the same irection. Inee, even in case a left-lane noe will be inclue into the route uring iscovery the communication will break shortly as soon as the left-lane noe will move out of the transmission range. Then, eventually the route will be establishe through the R noe an will be maintaine till the en of the experiment. In orer to evaluate the performance of in the presence of non-stable routes we specifie the scenario where the communication is performe between cars moving in opposite irections. In this way, the R car runs an FTP source proucing the ata estine to the L car. Initially R an L are locate with a two-hop istance. As soon as the cars move away of each other the route breaks requiring reiscovery. Consequently, the achieve throughput highly epens of the spee the cars move away of each other. The results presente in Fig. 8 show that,, an achieve similar throughput. + protocol emonstrates slight improvement ue to a better upate of routing metrics. The reason for ramatic throughput egraation with the car spee increase lies in frequent route failures ue to not stable network topology. In case the route failure is etecte all the above mentione protocols perform route iscovery flooing the network with route requests. As it is evience for the trace files, most of the frames transmitte over the network are RREQ an RREP. In orer to increase the performance of the routing protocol in highway scenario we performe a minor moification of + protocol which performs route iscovery only once in orer to etermine the coorinates an movement spee of the estination noe. Then, in case of a route breakage, instea of flooing the network with route requests, it forwars the packet to the noe locate closer to the estination assuming the coorinates of the estination i not change greatly from the last packet elivery. The route iscovery is being triggere only in case the estination noe move out from its previous location such that a forwaring noe locate within the transmission range of the estination

6 6 coorinates can not successfully eliver a packet. As a result, this moification, referre as + no RTS in Fig. 8, emonstrates a consierable throughput enhancement over other evaluate protocols. Base on the obtaine experimental results we conclue that in such an unstable by network topology but eterministic by movement patterns scenario as cars of a highway traitional routing protocols esigne for generic a-hoc networks o not perform well. However, their performance can be consierably improve fighting the roots of the problem like the reuction of network flooing base on the preiction of the car location. FTP Throughput (bps).5 x no RTS Spee (m/s) Fig. 8. FTP transfer between cars moving in the opposite irections. IV. CONCLUSIONS In this paper, a motion-base routing algorithm for vehicle a hoc networks,, is propose. The algorithm is completely istribute, since noes nee to communicate with only irect neighbors locate within their transmission range. Main feature of is the use of a routing metric which enables to exploit not only positioning information but also the irection the vehicles move: explicitly consiers ynamic changes in the network in aition to available topological information. Extensive evaluation outlines the avantages of, especially in case of high mobility of vehicles an frequent topology changes. In particular, a link layer integrate implementation of the protocol achieves goo performances in highway scenario with non-stable routes. For future work we consier an application of movement base routing metrics introuce in this paper to the set of greey routing protocols like GPSR []. We believe these protocols are probably the most promising from the performance benefits point of view in ynamic an highly unstable VANET topologies. Along with movement awareness we consier eveloping techniques which will exploit the peculiar properties of ifferent VANET scenarios (like the highway or the city gri). REFERENCES [] CarCar Communication Consortium. Conformance Guieline for Projects an Demonstrators. Preliminary Version. [] M. Bechler an L. Wolf. Mobility Management for Vehicular A Hoc Networks. Proc. of IEEE Vehicular Technology Conference 5, May 5. [] E.Royer an C.Toh. A Review of Current Routing Protocols for A Hoc Mobile Wireless Networks. IEEE Personal Communications, Apr. 999, pp [] M. Abolhasan, T. Wysocki an E. Dutkiewicz. A review of routing protocols for mobile a hoc networks. A Hoc Networks, Vol.,, pp. -. [5] S. Guo, O. Yang an Y. Shu. Improving Source Routing Reliability in Mobile A Hoc Networks. IEEE Transactions on Parallel an Distribute Systems, Vol. 6, Apr. 5, pp [6] O. H. Hussein, T. N. Saaawi an J. L. Myung Probability Routing Algorithm for Mobile A Hoc Networks Resources Management. IEEE Journal on Selecte Areas in Communications, Vol., Dec. 5, pp [7] S. Basagni, I. Chlamtac an V. R. Syrotiuk. A Distance Routing Effect Algorithm for Mobility (DREAM). Proc of ACM/IEEE International Conference on Mobile Computing an Networking 998, Dallas, TX, USA, 998, pp [8] Y.-B. Ko an N. H. Vaiya. Location-Aie Routing (LAR) in Mobile A Hoc Networks. ACM/Baltzer WINET J., Vol. 6,, pp. 7-. [9] M.Mauve, J.Wimer an H.Hartenstein. A Survey on Position-Base Routing in Mobile A Hoc Networks. IEEE Network, Nov./Dec., pp. -9. [] I.Stojmenovic. Position-Base Routing in A Hoc Networks. IEEE Communications Magazine, Vol., Jul., pp. 8-. [] H. Frey. Scalable Geographic Routing Algorithms for Wireless A Hoc Networks. IEEE Network, Vol. 8, Jul./Aug., pp. 8-. [] T. Park an K. G. Shin Optimal Traeoffs for Location-Base Routing in Large-Scale A Hoc Networks IEEE/ACM Transactions on Networking, Vol., Apr. 5, pp [] B. Karp an H. T. Kung, GPSR: Greey Perimeter Stateless Routing for Wireless Networks, in ACM/IEEE International Conference on Mobile Computing an Networking (MobiCom), Boston, MA, August,, pp. -5. [] G. Boato an F. Granelli. : a Movement-Base Algorithm for A Hoc Networks. Proc. of International Conference on Distribute Multimeia Systems, Sep.. [5] R. A. Santos, R. M. Ewars, A. Erwars Cluster-Base Location Routing Algorithm for Inter-Vehicle Communication. Proc. of IEEE Vehicular Technology Conference, Sep.. [6] L. Tonghong, S. K. Hazra an W. Seah A Position-Base Routing Protocol for Metropolitan Bus Networks. Proc. of IEEE Vehicular Technology Conference 5, 5. [7] W. Sun, H. Yamaguchi, K. Yukimasa an S. Kusumoto GVGri: A QoS Routing Protocol for Vehicular A Hoc Networks. Proc. of IEEE International Workshop on Quality of Service 6, 6. [8]

7 7 5 Spee: 5 units/secon Flooing 5 5 Spee: 5 units/secon Flooing Number of Messages 5 Number of messages Fig.. Comparison among, flooing an LAR in terms of routing overhea (nº. of messages) against transmission range, for ifferent values of the noe spee. 9 x 5 Spee: 5 units/secon 8 9 x 5 Spee: 5 units/secon 8 FTP throughput (bps) FTP throughput (bps) Fig.. throughput comparison for ifferent values of transmission range Range: units Range: units Spee (units/secon) Range: units Spee (units/secon) Range: 5 units Spee (units/secon) Spee (units/secon) Fig. 5. Performance comparison between, + an competing schemes in terms of performance against mobility, for ifferent transmission ranges.