Optimal overload response in sensor networks

Abstract

A single commodity network that models the information flow in an arbitrary topology sensor field that collects and forwards information to a backbone through certain designated gateway nodes is considered. Resilient operation in overload stress situations caused by unpredictable traffic or topology variations is investigated. That amounts to studying the network in instability mode, where the traffic load distribution is outside the throughput region. A fluid model is adopted where superflows model traffic forwarding and backlog formations at the network level. Quantitative performance metrics of the overload including throughput, lexicographic minimization, most balanced allocation, and amount of lost traffic due to buffer overflow are considered to capture the information loss process due to overflow in the network. Optimal superflows with respect to these metrics are characterized and a distributed asynchronous algorithm that computes such superflows is given. The characterization of the optimal superflow amounts to obtaining a structural decomposition of the network in a sequence of disjoint subregions with decreasing overload such that traffic flows only from regions of higher overload to regions of lower overload. The optimal superflow represents the smoothest trajectory to overflow, followed by the network in case of instability.