Effects of Partial Topology on Fault Diagnosis

Network components may experience faults for a variety of reasons, but it may not be immediately obvious which component failed. Fault diagnosis algorithms are required to localize failures and thereby enable the recovery process. Most current state of the art fault diagnosis algorithms assume full knowledge of the network topology, which may not be available in real scenarios. In this paper we examine the performance of one of these fault diagnosis algorithms, namely Max-Coverage (MC), when the topology is only partially known. We introduce a simple extension, called the Virtual Topology (VT), to correctly identify faults when a failure occurs in an unobserved component. We compare the performance of MC under partial topology knowledge with and without this extension to show that VT significantly improves correct diagnosis, but at the cost of a high number of false positives. Moreover, we demonstrate that correctly inferring areas of the unobserved network substantially mitigates the drawbacks associated with using VT.