Self-configurable fault monitoring in ad-hoc networks

While fault monitoring is a well-known issue in fixed wired networks, ad-hoc networks provide new challenges with respect to this issue: in particular, the monitoring task may be hindered at any time by the impossibility to observe a given node. This paper proposes a fault monitoring approach for ad-hoc networks which takes into account this constraint. Our approach is based on an information theory measure suitable to the intermittence of ad-hoc nodes and capable to detect network failures by inference. We define a distributed monitoring scheme with several collaborative detection methods, and we detail a self-configuration mechanism based on the K-means classification algorithm. We show how this solution can be integrated into a management architecture and evaluate its performance based an extensive set of simulations.