A Methodology Used to Optimize Probe Selection for Fault Localization

Due to the efficiency and adaptability, the active probing technique has become an attractive tool for fault localization in large and complex computer networks. It performs diagnosis by appropriately selecting the probes and analyzing the results. However, selecting an optimal probe set in such environment has been proven to be NP-hard problem. And, even the current approximate methods that can achieve near-optimal solutions have exponential computing time with the network size. To address this issue, we utilize the properties of conditional independence and directed-separation of Bayesian network, and propose a novel methodology which is used to estimate the approximate conditional independence of probes. According to the methodology, the model can be divided into several approximate independent subsets, on which the probes could be selected respectively. Furthermore, by integrating the methodology with a former representative probe selection algorithm which is called BPEA, we design a new efficient probe selection algorithm. Several experiments are given afterwards to show how our algorithm outperforms BPEA. And we also present that our algorithm can be used in large-scale computer networks while the former one can not. Moreover, the methodology can be applied to other probing based techniques as well.