Efficient Probe Prediction Algorithm for Fault Diagnosis in Computer Networks

Systems and networks management tasks such as fault diagnosis often require making quick and precise diagnosis to locate the root faults. Recently, the most popular method used to perform fault diagnosis is active probing. However, although active probing is an efficient tool for fault diagnosis, it has to send plenty of probes into the network, which has generated great traffic and imposed extra overload in the network. In order to address the problem, we propose a new approach based on probe prediction method by which we only need to measure a few probes in the probe set and predict out the responses of the rest probes. Then by applying Bayesian network into the diagnosis we quickly locate where the faults take place. The experiments have shown that the number of probes that needed to send over the network to perform diagnosis does not exceed 30% of the total probes, so the traffic generated by probes can be saved significantly.