Application of adaptive probing for fault diagnosis in computer networks.

摘要

Fault localization is the process of diagnosing the cause of failure from observed failure indications. With the increasing size and complexity of computer networks and with the increasing demands for quality of service guarantees in network services, fault localization has become an important network management task. Existing fault localization techniques are inadequate to address various challenges introduced by modern communication systems, such as the ability to perform reasoning in the presence of uncertainty about the underlying network information, diagnosis of multiple failures, low management traffic overhead, small deployment cost, quick diagnosis, and high accuracy.; This dissertation addresses these challenges by developing fault localization techniques using adaptive probing. Adaptive probing is a promising approach for performing efficient and effective fault localization. Adaptive probing sends test transactions over the network to diagnose the health of various network components. As probing involves overhead of sending management traffic in the network, adaptive probing reduces this overhead significantly by adapting the probe set to the observed network conditions, sending fewer probes in healthy areas and more probes in suspected areas of failure. To meet the challenges presented by modern communication systems, we provide adaptive probing solutions assuming the availability of deterministic as well as non-deterministic dependency information. We attempt to find multiple failures. Adaptive probing attempts to minimize the overhead of probe traffic by analyzing the network state from the previous probe results, and identifying the new probes to send that can give maximum information. Unlike the traditional technique of deploying passive monitors over a large part of a network, probing solutions reduce instrumentation overhead by requiring instrumentation of fewer nodes as probe stations. Also, we show through simulation results that adaptive probing provides a high detection ratio and a low false positive ratio.