Uncertainty Reasoning in Network Fault Diagnosis

The reliability and performance of overlay applications depend on the overlay network management capabilities to effectively diagnose and recover faults. Thus supporting accurate, robust and timely fault diagnosis is highly important for overlay applications to determine the root causes of faults even with noisy or minimum feedback from the network. This book presents novel techniques to improve the accuracy, robustness and performance of overlay fault diagnosis. First, this book develops a probabilistic reasoning framework that integrates passive and active approaches while avoiding their limitations. Second, this book designs a plausible reasoning framework to diagnose overlay faults and reason about the uncertainty when only end-user observations are available and network knowledge is very limited. Third, this book designs an optimal overlay distributed monitoring infrastructure for aggregating and delivering events with minimal latency and cost. Extensive evaluation via simulation and real-network experiments shows that the presented reasoning techniques significantly improve the efficiency and accuracy of overlay fault diagnosis