QoS in Vehicular and Intelligent Transport Networks Using Multipath Routing Christian Lazo Ramírez
Manuel Fernández Veiga
Instituto de Informática Universidad Austral de Chile Valdivia, Chile Email: [email protected]
Dep. de Enxeñería Telemática Universidad de Vigo Vigo, Spain Email: [email protected]
Abstract— Vehicular ad hoc networks (VANETs) consist of an spontaneous association of a group of vehicles that dynamically change their position and exchange data between each other, regarded as autonomous network segments with flat addressing schemes. However, its study has shown the benefits obtained by interconnecting them to fixed network segments and to the Internet. In this article the performance of a multipath routing protocol and its impact on global quality of service metrics will be analyzed by means of simulating different schemes of data transmision in a hybrid IPv6 VANET. The scenario consists of a fixed network segment with a central server and two gateways interconnecting two hierarchical VANET segment whose nodes show a high degree of mobility, such as vehicles in an urban environment.
I. I NTRODUCTION Vehicular Ad hoc NETworks (VANETs) , , ,  anticipate the technology for the future development of vehicular comumnications and intelligent transport systems.These networks cover many aplications related with safety on the road, warning drivers about accidents, congestions ahead the road or sending information to allow a server to centralize information from all vehicles about mechanical state, position, street status, emergency situations, etc. VANETs are formed by the spontaneous and uncoordinated association of a set of vehicles that change their positions dynamically, and that exchange data among them through wireless links. Such organization takes shape without aid nor participation of any external fixed network infrastructure. The routing protocols used in VANETs have as their main feature the ability to keep efficiently and reliably the communication between a pair of nodes (source-destination), even if the position and speed that they follow change quickly. Thus, when the communicating nodes are not directly connected, the information exchange is achieved by relaying packets through intermediate vehicles, or through the gateways linked to the network infrastructure. Currently, the behavior of these routing protocols is an area of active research. Most of them are based on enhanced versions of well known unicast routing protocols standardized by the IETF (Internet Engineering Task Force), such as OLSR , AODV  or DSR . Routing protocols used in VANETs include the proper mechanisms to exchange data packets within the network
1-4244-0755-9/07/$20.00 '2007 IEEE
range, thereby allowing vehicle-to-vehicle (V-V) communications. But, in order to send packets toward external networks or to a wide area network infrastructure, as Internet or other fixed network, it becomes necessary the collaboration of some network element to serve as a gateway able to provide the functions that will set up the connection from the moving vehicle toward the fixed infrastructure (or V-I communications). To offer this kind of connectivity, the device serving as a gateway must use wireless transmission toward the VANET and another access link to the fixed network , . The gateway used to interconnect both network segments, fixed and mobile, can operate in three modes, namely, active, reactive or hybrid, depending on the network protocols used. However, the impact that each of the modes causes on performance is still unclear , , since performance is also influenced by other external factors, e.g., the degree of mobility of the vehicles, the distance in number of hops between the gateway and the vehicle participating in the communication, the routing protocol used inside the VANET, or the degree of partitioning in the VANET, among others. This paper quantifies, through a simulation study, the overall improvements in performance (especially in QoS related metrics) that arise in a next generation vehicular network due to the use of a node-disjoint multipath routing protocol (MRP). The network traffic is composed of data transmissions between vehicles in each VANET, in form of V-V connections, and of transmissions from the vehicles to the infrastructure (fixed) network in an V-I scheme. The rest of the paper is structured as follows. In Section II the different connection strategies used in the gateways to provide Internet connectivity to the VANET are presented. Section III gives an overview of the mobility models commonly used in this type of networks. Section IV describes the details of the multipath routing protocol used in this setting, and Section V states the network assumptions on which the simulations are based. In Section VI we present and discuss the results, and in Section VII we summarize the conclusions of this work. II. C ONNECTION S CHEMES Though VANETs were, at first, conceived as a solution for isolated vehicular networks, soon it was evident that there were
greater potentialities and an increased number of applications in connecting VANETs to fixed networks, i.e., to Internet. Ad hoc networks connected to fixed networks are also known as hybrid networks . In this approach, one or more nodes in the VANET are also part of the fixed network segment through some of their interfaces. So, it is this kind of nodes (the gateways) who provide the connectivity to Internet. For a node in the VANET to be able to send or receive traffic from Internet, it must be informed which are the gateways available to forward the packets. There exist three approaches to fulfill this task . A. Proactive Gateway In this case, the gateway periodically floods the network with messages announcing its own presence and availability. Hence, the vehicles know permanently who is the gateway that can trust to forward their outbound packets. It is clear that, with his approach, the latency is low but there is a high consumption of bandwidth inside the VANET. B. Reactive Gateway In this solution the gateways play a passive role, before sending the first packet to the fixed network, a node in the VANET requests the identity of the gateways. This procedure is repeated periodically afterwards. C. Hybrid Gateway A third possibility is the hybrid gateway. In this solution, a gateway only informs proactively of its presence to the vehicles within its coverage radius, i.e., to the vehicles one hop away. Farther vehicles get the status information using the reactive method described above. It is important to remark that, regardless of the connection method offered by the gateway, all the traffic that traverses it is encapsulated into tunnels to enable the forwarding of packets originated in the VANET across the infrastructure network. III. M OBILITY M ODELS When considering mobility models, one should take into account that the shape, speed and trajectory followed by the mobiles directly determine the degree of partitioning and the connectivity level achieved in the network. Some measures of interest for these variables are the following: • Network partitioning. For a given set of vehicles, and for a specified period of time, it is the number of actual network segments. • Spatial dependence. The measure of dependence between vehicles movements. So,if two mobiles have similar speed and the same direction, then they have high spatial dependence. • Temporal dependence. The measure of dependence between two mobiles when their instantaneous speed is observed in magnitude and direction. Given their features, we can stress that VANETs are useful to provide connectivity to different application contexts. Hence, each of those scenarios should be represented with
the most realistic model so as the simulation studies yield meaningful results. Among the main mobility models used by the research community in vehicular networks, we outline the following . A. RPGM This mobility model for groups is mainly used to simulate battlefield communications. In such context, the group leader moves at a given speed and the rest of the group follows the leader’s movement at a similar speed and direction. Both the speed and direction are periodically adjusted to the changes made by the group’s leader. So, this model exhibits a high spatial dependence. In the context of vehicular networks, the model approximates the behavior of a row of mobiles on road. B. Freeway/Highway Freeway/highway models are useful to assess the behavior in VANETs in long distance high-capacity roads. Three levels of speed are defined here (slow, medium, and fast), representing the lanes in the freeway. Mobiles can change their lane, and consequently their speed. The change is made gradually, i.e., to shift from the slow to the fast lane, a vehicle must pass first to the medium speed. C. Manhattan Grid The Manhattan Grid mobility model is particularly suited to assess the performance of VANETs in urban environments. The model allows to capture accurately the behavior of moving vehicles across the streets of a virtual city. To this end, a grid is defined where every row and column represent streets and intersections. In such model, the mobiles can move randomly at a given average speed, they can stop and turn their direction (leftwards or rightwards), or can continue along the same street (see Fig. 1). The Manhattan Grid model imposes severe geographical restrictions to the movements of the vehicles, thus increasing the probability of network partitioning. On the other side, one of its advantages is to allow simulations with a good degree of approximation to the situation found in real urban
Street structure with Manhattan Grid.
environments. Moreover, due to the rectangular movement of vehicles, the Manhattan Grid shows a high degree of both spatial and temporal correlation. IV. ROUTING IN VANET S Many routing protocols for ad hoc networks have been proposed and developed in the last years , , , , . Furthermore, there have been conducted a number of comparisons between the different proposals , , . However, the common view is that reactive protocols show superior performance, so they have gained better acceptance within the IETF, particularly the AODV (Ad hoc On demand Distance Vector) . This protocol has also been extended so as to compute and use multiple paths inside the VANET , as well as to bring Internet connectivity to the domain of VANETs . In this last work, the multipath version of AODV (AODVM) is evaluated for the case of hybrid networks using gateways. Let us give a brief overview of AODVM . A. Ad hoc On Demand Distance Vector Multipath Routing The proposal of AODVM consists of an extension of the well-known reactive Ad-Hoc routing protocol AODV. AODV is based on a distance vector algorithm for computing routes. It avoids the cases of slow convergence (including the problem of count to infinity), it detects routing loops, and, finally, it does not discard duplicate route request packets (RREQ). The latter feature is crucial to discover node-disjoint multiple paths. The procedure to discover a route in AODVM is started when a source node requests a route toward a destination. The source node cannot find the route in its routing table, so it increases a sequence number and broadcasts a RREQ message. Neighbor nodes that receive the RREQ record the address of the origin node that sent the request, the address of the destination node, and the address of its neighbor node that relayed the request. Additionally, the node creates or updates the reverse routes toward the origin node of the route request message, with new sequence numbers. Eventually, when the destination receives the first route request message from some of its neighbor nodes, it updates the sequence number and generates a route reply (RREP) message of response with a new field, Last hop ID (LHID). LHID contains obviously the address of the last node in the chain of relay nodes toward de destination. Then, the response packet conveying this field is sent backwards to the origin node, following the same route that the request message but in reverse order. The same procedure is repeated for each of the route request messages arriving from a different neighbor, so that all the responses differ in the values conveyed in the LHID field (recall that the algorithm searches for node disjoint paths). In order to discover these node-disjoint paths, the intermediate nodes can only participate in a single route for every origin-destination pair. Thus, only the shortest route in their routing table satisfying the connectivity constraints is kept, while the rest are deleted.
In AODVM, the origin node must confirm the discovery of routes reported by the response messages by sending an explicit route confirmation message (RRCM). The RRCM is appended (piggybacked) to the first data packet sent along such route. Note that the confirmation message informs all the intermediate routes of the validity of that path. The active paths between origin and destination is also kept alive by sending periodically HELLO messages or modified response messages with TTL set to one. In any case, the messages are only received by the direct neighbors, acting as an indication to refresh the routing table and avoid the expiration of its entries. Accordingly, after a prescribed time no HELLO messages are received from a neighbor, the node assumes that the device is unavailable and discards the route. V. N ETWORK A SSUMPTIONS We report in the following a simulation study of the behavior and performance of the multipath routing protocol for a vehicular network supporting data transmissions directly between mobiles (V-V) and between the fixed network segment and the VANET network, via the gateway nodes. The simulations pay special attention to the impact of the multipath routes on the fundamental QoS metrics (delay, packet delivery fraction). All the experiments have been conducted with the ns tool . Before discussing the results, in this section we will state the network model and the parameters used in the experiments. A. Simulation Model The network setting consists of 4 nodes in a fixed network segment. The number of vehicles in the VANET was variable, between 18 and 68, and they were split into two segments. All the nodes employ AODVM as routing protocol, and both domains, VANET and infrastructure network, were connected by means of two gateways, as depicted in Fig. 2. The topology of the VANET is distributed over a 1200 m × 600 m area, divided into 7 × 5 streets as depicted in Fig. 1. The two gateways operate in reactive mode and are fixed at coordinates 310, 330) and (910, 250). The mobiles move
TABLE II C ONFIGURATION PARAMETERS OF TRAFFIC .
randomly according to the Manhattan Grid model, with an average speed of 18 Km/h. Both the vehicles and the gateways have an omnidirectional antenna with a transmission range equal to 250 m. The medium access protocol is IEEE 802.11, and the nodes in the infrastructurenetworks and the gateways are connected with bidirectional links with 100 Mb/s capacity and 2 ms delay. A summary of these configuration parameter appears in Table I.
Source All Vehicle Vehicle Vehicle Vehicle
Nodes Destination Server 1 6 Vehicle 14 15 Vehicle 23 10 Vehicle 19 20 Vehicle 11
Data packets size 512 kb 2.5 s 256 kb 0.2 s 256 kb 0.2 s 256 kb 0.2 s 256 kb 0.2 s
Start of tx. 0