Veer: A Trajectory-Based Peer Selection Algorithm for Networks of Vehicles S´ avio Rodrigues Cavalcanti ⋆ , Miguel Elias Mitre Campista ⋆ , Fehmi Ben Abdesslem ⋄ , Lu´ ıs Henrique Maciel Kosmalski Costa ⋆ , Marcelo Dias de Amorim ⋄ , and Otto Carlos Muniz Bandeira Duarte ⋆ ⋆ COPPE/POLI – Universidade Federal do Rio de Janeiro, Brazil ⋄ LIP6/CNRS – UPMC Univ Paris 06, France Emails: {savio,miguel,luish,otto}@gta.ufrj.br ; {fehmi,amorim}@rp.lip6.fr. Abstract—We propose Veer, a peer selection algorithm for file exchange in vehicular ad hoc networks. The innovative aspect of Veer is that it relies on embedded GPS devices to determine individual mobility patterns and, consequently, to help select the peers that increase the probability that a file transfer succeeds. Transfer opportunities are computed in advance based on an esti- mation of the contact time window between a node and its selected peers. We show that such an opportunistic one-hop transfer mechanism leads to better results than traditional multi-hop file transfers both in terms of goodput and delivery ratio. I. I NTRODUCTION One of the challenging goals of vehicular ad hoc networks (VANETs) is the support of legacy applica- tions such as peer-to-peer file sharing. In file sharing applications, the contact time between two peers must be long enough so that they can establish a connection and transfer the required amount of data. The difficulty here is that file sizes might vary within a wide range of values – this implies that we must first evaluate if a contact will be long enough to support a transfer. Note that in the particular context of wireless networking, such a contact time can be very short, as it refers to the time peers are within the communication range of each other. The contact time depends on many parameters, such as the relative speed of the peers, the wireless technology used, or the environmental conditions [1]. On the other hand, if communication protocols are properly tuned, even very short contact times can be used to transfer a few megabytes of data, using off-the-shelf wireless equipment [2]. Routing protocols for VANETs have been proposed in the literature, but in general they rely on exist- ing solutions issued from research on mobile ad hoc networks (MANET) [3], [4]. These protocols present interesting properties (e.g., management of disconnected scenarios), but face some scalability problems in highly dynamic situations. With proactive protocols, a storm of control messages is required in order to maintain routing tables. Although reactive protocols are preferred in highly dynamic scenarios, breaking paths causes ex- cessive broadcasting for new routes to be discovered. Because of the potentially short contact windows and the specificities of file sharing applications, relying on existing solutions is not sufficient. For example, as a source does not know the contact duration, a large-file transfer may be abruptly canceled. More specifically, we need a solution that tackles the two following goals at the same time: • Expected contact time. The strategy used by a peer to transfer data to a neighbor should rely on the expected time window these two peers remain in contact. On the one hand, too short contacts might be insufficient or should be used only for trans- ferring small packets. On the other hand, longer contacts are of course more appropriate for larger messages. Having the possibility to estimate future contacts would be of great utility to properly sched- ule transfers and thus increase the overall capacity of the system. • Avoid multihop communication. Finding a long enough contact is by itself a difficult task, espe- cially in highly dynamic networks such as VANETs. The use of multihop-based routing protocols would require a product of contacts that would depend on the path length (in number of hops). We advo- cate, whenever possible, the avoidance of multihop communications; instead, nodes should rely on their