Research On The Key Techniques For Intelligent Network Fault Diagnosis

Fault diagnosis is a central aspect of network fault management. It is important to make a fault diagnosis rapidly and accurately for enhancing the robustness, reliability and service availability in network system. Meanwhile, it is significant to improve the capability of fault diagnosis for reducing failure repair time and decreasing the cost of maintaining the network.At present, a lot of research work has been done in the fault diagnosis area and many technologies are proposed. However, there are still many issues remain to be resolved, such as the network diagnosis problem under the circumstances of a great deal of alarm loss and spuriousness alarms and the imprecise fault model. Since this phenomenon is common in the current network fault management, it is very necessary to find an effective solution for solving the above problem.Aiming to achieve accurate and efficient intelligent network fault diagnosis, the dissertation focuses on the research of alarm information in the fault management and fault diagnosis problems when the uncertainty of fault models and localization is high. The dissertation emphasizes particularly on the creation of fault propagation models, the analysis and improvement of fault localization algorithms, the test scheme in the uncertain condition and the design of services-oriented architecture for the network fault management system. The main contents of the dissertation are as follows:1. Based on the comparison of current fault propagation models (FPM), bipartite graph model is adopted as fault propagation model, of which the formal definition is given. Through a top-down and hierarchical way, we divide service-oriented networks fault management tasks into separated layers and the notion of diagnosis by layer is provided. Finally the methods of fault propagation modeling for each separated layers are proposed.2. Taking false faults into consideration, we present an improved bipartite graph fault propagation model. Through the transformation of the target function of the fault localization problem, the localization problem is transformed to aproblem of 0-1 program. Largarian Relaxation Algorithm (LRA) is adopted to solve the problem of 0-1 program and the effectiveness of the algorithm is verified through experiments, which show that LRA performs a little better than IHU both in precision and efficiency.3. Considering the factor of alarm loss, a symptom node layer is added to the improved bipartite graph model and a network fault propagation model using the three-layer belief network is proposed. According to this module, we present a space-search-based fault localization method. It first heuristically searches the potential fault space to obtain a fault hypotheses subspace that satisfies the correlation and non-redundancy constraints. The hypothesis with the largest confidence is chosen as the solution. The heuristic search algorithms include a refined IHU algorithm (RIHU) and a recursive minimum hypothesis creating algorithm (RHC). Experiments show that the above algorithms not only enhance the detection rate but also decrease the false positive rate. In the case of a great many of alarm loss and false alarms, our algorithms perform better than LRA.4. Because the uncertainty factors in network fault propagation model and the results of fault localization algorithm often cause the inaccurateness of the final diagnosis results, we propose a fault diagnosis solution which based on the L-best fault hypotheses. The solution first uses fault localization algorithm to obtain the L-best fault hypotheses. Then the L-best fault hypotheses are tested step by step, of which the results are used to find the actual faults in the network. The testing strategies include information-gain-based technique and mutation-information-based techniques.5. On the basis of the research of the key techniques for the network fault diagnosis, we design Service-Oriented Architecture for Network Fault Management System, and describe the detailed design of two main function modules—Data Collection Subsystem and Fault Diagnosis Subsystem.