Threshold-based Policies in Mobile Infostation Networks Furuzan Atay & Christopher Rose WINLAB, Rutgers University 73 Brett Rd.,Piscataway, NJ 08854-8060 Email: furuzan/crose@winlab.rutgers.edu Abstract— Mobility can increase throughput in ad hoc net- works by providing channel variation. If delay constraints are very loose, it is possible for a given packet to observe many different network topologies as nodes move relative one another, and these different topologies can be treated as diversity. Oppor- tunistic strategies can exploit large scale changes in the channel quality to achieve higher throughput. However, to attain these higher capacities, delays on the order of the node mobility time constants must be tolerated. In this paper, we design cost functions based on delay and transmit power and implement simple greedy cost-minimizing strategies to enable the trade-off between mobility, delay and throughput. In particular, we study the performance of networks where all the packets are routed using simple threshold rules. We also examine the scaling properties of throughput and delay of our strategies. I. I NTRODUCTION Recently, it has been shown that for networks of geo- graphically fixed nodes throughput capacity is ”not scalable”. That is, in the limit of large numbers of nodes, throughput capacity per node goes to zero. Even if transmission ranges and transmission schedules are chosen optimally, as the number of nodes increases, throughput capacity per node decreases as 1/ √ N where N is the number of nodes. [1] One new line of thought involves node mobility – previously seen as undesirable since it complicates routing and can cause packet loss owing to intermittent node connectivity. However, if packets have loose delay constraints, mobility can not only increase throughput but also make capacity per user scalable [2]. The key idea is limiting both the number of hops 1 and the average hop length simultaneously regardless of network size – an impossibility in any fixed network under uniform traffic assumptions. This new idea constitutes the theoretical motivation behind the ”mobile infostation architecture” [3]. In mobile infostations a packet can travel between nodes when the conditions are favorable and sojourn at a relay node in the meantime. That is, relay nodes need not forward a packet as soon as it is received. If the next hop is too costly (in terms of some suitable network resource), packets can be retained until the next good transmission opportunity. It is possible for a given packet to observe many different network topologies as nodes move relative one another, and these different topologies can be treated as diversity. In generic ad hoc networks, many applications have strict delay constraints requiring the network to be connected most of the time. A mobile infostation architecture, on the other hand, targets applications with loose delay constraints and high data rate requirements. Thus, intermittent connectivity is both tolerated and expected. 1 In the case of [2], the number of hops was limited to two. In this paper, we study the trade-off between throughput and delay in mobile infostations. We define cost functions based on transmit power and delay, and implement simple greedy optimizations at packet level. In [4], different packet-oriented strategies have been studied and it has been concluded that from a practical and computational perspective, threshold rules are the best candidates to be used in network studies in which packets interact. Thus, we base our protocol on threshold rules. II. RELATED WORK In [3], the authors studied the throughput performance of mobile infostations by accepting a potentially large delay. The optimal transmission ranges to maximize local throughput is calculated and it was shown that the optimal transmission range of mobile infostations is much shorter than (5 to 10 times) that of generic ad hoc networks. Lately, there have been attempts to improve the delays of architectures that rely on mobility [5]–[7]. One way of decreasing the packet delay is generating and distributing more than one copy of the same packet [5] and [7]. Another way is to allow each packet to make more hops than required 2 . [6] considers a multiple hop approach to alleviate delay. However, the proposed protocol requires a two-tier hierarchical architecture where the sources and destinations are stable and the relays are mobile. It also assumes that all mobile nodes know their future trajectories for random time and they share this knowledge with each other and stable nodes. In this work, we consider a homogenous network where all the nodes are mobile. We assume that nodes can obtain up- dated topology information (not necessarily global topology) after each hop, but does not know future topologies. The rest of this paper is organized as follows: First, we will introduce the basic model and the cost structure. In section-IV we will describe the threshold based packet policy. A multiple packet simulation model will be described in section-V. In section-VI we will examine the results. III. MODEL AND ASSUMPTIONS We consider a network of packets that interact and compete with each other for network resources. The objective is to achieve low delay and high throughput. We define a cost func- tion reflecting these objectives and evaluate optimal operating points by varying the weight of cost components. We assume discrete intervals of duration δ during which packets can move directly between two nodes or stay put to await more favorable conditions. Time is measured in integer 2 The minimum number of hops required to achieve a scalable throughput depends on the system details like the traffic model. In the case of [2] at least 2 hops are required.