Next-Hop Decision-Making in Mobility-Controlled Message Ferrying Networks

Abstract

Message ferries support delay-tolerant networking in scenarios where nodes are too far from each other to communicate directly. We study scenarios where a data ferry moves between a fixed set of stationary nodes not connected otherwise. Thus, ferrying is assumed to be the only means of communication in the network. In order to minimize the average message delivery time, the ferry dynamically decides on the next node to visit. For this, we assume local knowledge only, i.e. the decision made by the ferry is solely based on locally derived information. Our studies are based on an abstract state transition model for the system. Based on this model we analyze the performance of different algorithms. This includes a static planning algorithm selecting the shortest Euler tour (TSP) and our dynamic decision algorithm (SOFCOM) based on local knowledge. In order to study the performance of both algorithms, we compare these with results derived by an idealized nondeterministic algorithm (oracle) assuming global as well as future knowledge on the generated messages. Our studies are based on symmetric as well as asymmetric traffic with exponential message arrivals. We show that our SOFCOM algorithm typically outperforms the TSP algorithm and that it achieves results close to the oracle solution. Especially important, our studies show that the benefits of global knowledge of the system state are rather small, and that local knowledge is sufficient to achieve very good results.