Cross-domain network fault localization

Abstract

Prior research has focused on intra-domain fault localization leaving the cross-domain problem largely unaddressed. Faults often have widespread effects, which if correlated, could significantly improve fault localization. For both competitive and security reasons, domain managers hesitate to share fault observations even when doing so may significantly ease fault localization. This dissertation presents a charac- terization of the problem space in terms of inference accuracy, privacy, and scalability, and provides a framework to evaluate any design in the design spectrum. This frame-work not only explicitly models the inference accuracy and privacy requirements for discussing and reasoning over cross-domain problems, but also addresses scalability impacts and facilitates the re-use of existing fault localization algorithms while enforcing domain privacy policies. The dissertation provides a graph-digest-based approach with which participating network domains can exchange abstracted graphs that represent network fault propagation models. The research explores feasibility of this approach via implementation of an inference graph-based design in a cross-domain network setting. The results show a substantial improvement in cross-domain fault localization accuracy and inference speed by using the inference-graph-digest based approach.