Fault Diagnosis in IP Networks via Multicast Probing: Noisy Measurements

In this paper, we address network fault diagnosis using multicast-based probing. The problem is inspired by and relevant to multicast-based IPTV services being heavily deployed by telecom operators around the world. Specifically, we introduce a disjunctive fault model for multicast-based probing wherein a receiver probe records a faulty measurement if and only if at least at least one of its ancestor nodes is in a faulty state. The model also allows for the presence of measurement errors. Using this fault model, we derive key results concerning the most probable fault scenario nd the most likely fault given the observed probe values. We consider both noiseless screnarios and noisy scenarios that are suscepticle to measurement errors. Our procesures exploit the underlying structure of the multicast tree and are significantly more efficient than generic computational procedures for probabilistic inference. They can form the basis for accurate and timely fault diagnosis and service quality resolution procedures as part of performance management and customer care systems respectively.