An Application to Two-Hop Forwarding of a Model of Buffer Occupancy in ICNs Marco Cello DITEN University of Genoa, Italy marco.cello@unige.it Giorgio Gnecco DIBRIS University of Genoa, Italy giorgio.gnecco@dist.unige.it Mario Marchese DITEN University of Genoa, Italy mario.marchese@unige.it Marcello Sanguineti DIBRIS University of Genoa, Italy marcello.sanguineti@unige.it Abstract – An application of the model proposed in Cello at al., A Model of Buffer Occupancy in ICNs, IEEE Com- munications Letters, to appear is investigated. Such a model provides a relationship in the z -domain between the discrete probability densities of the buffer state occupancies of the nodes in the network and the sizes of the arriving bulks. Un- der a class of two-hop forwarding strategies, expressions are obtained for the average buffer occupancy and its standard deviation. Keywords: Intermittently-connected networks, congestion control, ad-hoc networks, Markov chains, epidemic routing. 1 Introduction In the last years various applications emerged, where net- works operate under conditions in which the assumptions of “universal connectivity” and “global information” do not hold. Examples are sensor networks [8], social networks or pocket switched networks [6], smart environments, and ve- hicular ad-hoc networks [18]. A common denomination of such contexts is Intermittently Connected Networks (ICNs). As in such contexts the networks may be disconnected most of the time or it may even happen that there is never an end-to-end path available between source and a destination, classical routing and data delivery-approaches (see, e.g., [1]) fail [14]. In [12] it was proved that the expected throughput of re- active protocols (which compute a route only when it is needed) is connected with the average path duration pd, the time to repair a broken path tr, and the source data rate r through the relationship: throughput = max(0,r(1 − tr pd )). However, node mobility leads to frequent disconnections, thus reducing the average path duration significantly. Con- sequently, in most cases tr is expected to be larger than the path duration, which implies that the expected throughput is close to zero. Other approaches to deal with routing in ICNs involve the use of additional communications resources (e.g., satellite, UAV, message ferries) forced to follow a given tra- jectory between disconnected parts of the network, in order to bridge the gap [10, 20] (DataMule, Message Ferries, etc.). In other cases, such as in inter-planetary networks [2], in- termittent connectivity is predictable, so classical routing al- gorithms may be adapted to compute shortest delivery time paths by taking into account future connectivity [7]. Often, neither additional resources with controlled behav- ior nor predictable trajectories are available. In such cases, one of the most common approaches is epidemic routing [17], which is based on the replication and transmission of messages to newly-discovered contacts that do not already possess a copy of the message. In epidemic routing each node maintains a buffer, consisting of messages that it has originated and messages that it is buffering on behalf of other nodes. When two nodes meet each other, they decide how many and which stored messages are exchanged. In turn, each node requests copies of messages from the other. In the simplest case, epidemic routing is flooding: each time a con- tact happens, all messages that are not in common between the two nodes are replicated. In general, however, message replication performed by epidemic routing paradigms imposes a high storage overhead on wireless nodes [19] and very likely node buffers run out of capacity. More sophisticated techniques can be used to limit the number of message transfers. Existing epidemic protocols try to avoid congestion by limiting, either in a de- terministic [13] or in a non-deterministic way [11, 16], the number of copies of a message inside the network. So, an analytical framework for congestion control management is needed. This is the subject of the present contribution, which ex- tends our previous work [3] by applying the model proposed therein to the kind of epidemic routing known as two-hop forwarding. First, we describe the analytical framework de- veloped in [3], based on bulk arrival and bulk service queues, to model ICN nodes behavior (Section 2). Then we discuss a Proc. of the 2012 7th International Conference on System of Systems Engineering, Genoa, Italy - 16-19 July 2012 491