Research Of Distributed Network Fault Management

With the expansion of modern networks, fault management of network faces challenge caused by the increasing complexity of networks, which brings out new requirements for network fault management both in research work and in application area. Under the traditional manager/agent model of centralized network management, the manager side collects almost all the management functions. On the other hand, agent seldom participates management. So managers have to interchange information with devices located at distribution, which results in degrading in network performance. What's more, the activities of manager and agent are designed as fixed paradigm. Thus some inherent known problems of centralization framework are the lack of scalability, reliability, flexibility and strong local processing ability.To satisfy these new requirements, a distributed network fault management framework based on multi-agent system has been proposed in this thesis. It takes full advantage of agent's distribution and cooperation to overcome drawbacks of centralized network management. It gives solutions to realize distribution and automation of network fault management. Network is divided into several management parts and each part is assigned a group of managing agents. The agent group performs managing tasks in local network part. When needed, several agent groups cooperate to resolve complex managing tasks. We also design the structure and functions of four kinds of agents, which perform different managing tasks including fault detection, location and diagnosis.Then according to the process of fault management, this thesis introduces implementation of agent's management functions in different phase of fault management. In fault detection, detection agent captures abundant information about network situation. After information obtained, detection agent deal with such information, including aggregation of packet information and dynamic isolation of fault symptoms. Following fault detection, location and diagnosis agent is spur to initialization. Fault location process is based on distributed System-Level-Diagnosis algorithm. Many location and diagnosis agents cooperate as System-Level-Diagnosis algorithm directed in order to find the malfunction element. We propose two-level test algorithm based on ADSD and Hi-ADSD algorithm. The proposed algorithm improves ADSD algorithm on aspect of diagnosis latency. Diagnosing of the malfunction element tells the type and reason of fault. We sum up rules of fault types corresponding to fault symptoms. And parts of diagnosing program are provided.