A traffic-aware infrastructurebased Architecture for intervehicules File Sharing Amine K., Sadeq Ali M., Mesut G. 2008 IEEE 學生:羅英辰、董藝興 學號:M97G0216、M98G0106 1 Introduction Recent studies have shown in which extent peer-to-peer file sharing traffic occurs in the global Internet. Vehicular ad hoc networks (VANET) represent an emerging environment that illustrates the combination of these trends. 2 The rest of this paper is organized as follows. Firstly we present the proposed overlay organization of access points and subsequently we propose a clustering strategy for that organization. Finally, we give an overview of the protocol that has been implemented to support the architecture. 3 Architecture The main goal in the design of the architecture was to provide searching for shared files not only among vehicles during opportunistic meetings, but also among all vehicles in the VANET. The main scenario assumed by our architecture is illustrated by Fig. 1. Fig.1 uses CarTorrent protocol. 4 In this scenario car 1 is looking for a file that is shared by car 2. Use access points to avoid some important restriction. 5 To fulfill this requirements, it is required to build and manage shared-file indexes in a distributed way in order to avoid bottlenecks. We will show the importance of synchronizing concurrent requests and deliveries to reduce as much as possible the overhead and decrease the download time of files. 6 Architecture - Overlay Protocol To use P2P overlay protocols such as Chord, is a good way for implementing vehicular network and the high mobility of the nodes demand for a distributed management of filerequests. Chord uses consistent hashing to map nodes onto an m-bit circular identifier space. 7 A thorough analysis of the overhead has shown that the standard Chord-model with a single ring would result in heavy message-overhead, especially for the updating of car-positions. As it was clear that most message-traffic will be caused by the management for the car-positions, we decided to migrate this functionality to n the super nodes (clusterheads), at the same time keeping the file-management decentralized. 8 9 Architecture - Clustering A key point in the architecture is to build clusters and choose supernodes. To build clusters in such a way that intercluster vehicle-traffic is minimized, as it will be reflected through an inter-supernode message overhead. Clusters have the most possible maximum coherence and the most possible minimum coupling to other clusters. 10 We decided to build clusters using the k Medoid clustering algorithm. • • i=1…n, j=1…k Cp is the pth cluster such that d(xi,mp) ≤ d(xi,mj) with j = 1. . . k and i = 1. . . n. 11 The algorithm could be resumed in the following four steps: 1. 2. 3. 4. Choose k points to be the initial cluster-heads (Medoids). Assign each node to the closest Medoid. When all nodes have been assigned, try swapping cluster-nodes with their cluster-heads and see if the costs are decreased. Repeat Steps 2 and 3 until the Medoids no longer move. 12 13 The coherence of a cluster, in our case, is measured by the traffic of vehicles inside it rather than distances between its nodes. Let σst = σts denote the number of shortest paths from s ∈ V to t ∈V , where σss = 1 by convention. The following measure denotes the betweenness CB(v) for vertex v as defined by Freeman . 14 Architecture-Protocol Details It is clear that with the changes made to the Chord protocol, we have to adjust the basic functionalities of this protocol to cope with the splitting of file-indexes between clusters and the introduction of clusterheads in their organization. 15 Protocol Details - Synchronization of requests The schema of synchronization deployed therefore can be resumed in four rules: 1. 2. 3. 4. Requests for each file have to be temporarily registered in the corresponding file-index. As soon as a gateway retrieves a file, it sends a message to clear the list of requests registered for this file (or check whether it has already been cleared). If the requests-list is empty this gateway abandons the delivery of this file otherwise it takes into responsibility to transmit the retrieved file to all requesting cars. From time to time, gateways have to check whether the non yet retrieved files have been already retrieved by another gateway by checking the requests-cache. 16 In this context of dissemination of fully autonomous delivery-messages, we introduced the notion of responsibility-based (prioritybased) delivery process to be able to control and manage the refreshing process of deliveries in a decentralized way. 17 Protocol Details - Prediction of NextCar-Positions The prediction of next car positions is used in: First stage to avoid flooding access points with filerequests and hence quickly retrieve the sought-after files. Second stage to accelerate the deliverance of files to cars by forwarding the retrieved files to probable next-positions of requesting cars. 18 Simulation We have defined a grid of 5 blocks at both the horizontal and the vertical axis, where each block represents 1 km. Cars have a speed of about 50km/h with a turn probability of 0.7. The number of access points has been varied from 13 to 23 and thereof 2 to 6 supernodes should be chosen.(Fig. 3b) 19 All nodes use IEEE 802.11b MAC operating at 5 Mbps. The transmission range is about 250 m. The radio model has been specied according to the results of the drive-thru-measurements. 20 Performance Evaluation In our performance evaluation, we have focused on two aspects. The first aspect is the clustering strategy. Protocol Details. 21 22 23 24 Conclusions and Future Work Using a suitable clustering strategy, we could prove that taking into account the form of car-traffic has a direct influence on reducing the overhead and the download time of files. We want also to enhance the prediction strategy for car-movements by identifying redundant paths in cartraffics. On the mobile side, we are developing a protocol that uses our architecture to support a direct exchange between cars as it is the case in the CarTorrent project. 25