Research And Application Of Intelligent Network Fault Diagnosis

With the fast expansion of network-scale and the constant increase of heterogeneous network, the importance of network fault diagnosis was obvious specially. After analyzing some approaches of alarm correlation, this thesis proposed a prototype system of network fault diagnosis that had been elementarily validated on school net, which consisted of Rule Reasoning Module, Dependency Graph Reasoning Module and Data Mining Module. Then it emphatically expatiated on a general method that set up alarm correlation tool and provided an algorithm of heuristic fault localization based on dependency graphs -Divide and Conquer Maximum Mutual Dependency Algorithm. And it discussed the problem that occurred when the algorithm was executed in condition of independent failure. This thesis improved the algorithm. The simulated experiment result shows that the improved algorithm decreases computational complexity. But in the case that the algorithms fail to find the optimum solution, the divide and conquer algorithm is more accurate.