Non-intrusive Neighbor Prediction in Sparse MANETs

To increase the availability of mission critical services and information in sparse MANETs with frequent and/or long term network partitions, we aim to develop efficient replication and placement algorithms. The prediction of the neighborhood of a node is one core element in these algorithms. In this paper we present a neighborhood prediction algorithm based on the Sequential Monte Carlo framework, i.e., recursive Bayesian filters using a set of random samples, which are updated and propagated by the filter. The algorithm works without location information and extracts only information from the local routing table to predict the future neighborhood of the node. We have performed extensive experiments to evaluate the accuracy of the prediction algorithm. The predicted connection and disconnection times follow closely the "true" distribution as registered by the routing protocol.