Study Of The Hierarchical And Distributed Network Fault Management

The network fault management is the basement work and the most important task in the network management which mainly includes such key technologies as the network fault detection, the network fault localization, the network fault isolation and the network fault restores. Fault management deals with how to guarantee to get the network monitoring data efficiently, how to optimize the network fault management policy, how to fast locate the network fault and so on. In this thesis, we discuss the network fault management from four aspects: the algorithm of network fault monitoring, the policy optimization of the network fault location, the fast diagnosis algorithm of the network fault and the network fault management application in Ad hoc network. The main contributions of this thesis are listed as follows:1. In the aspect of network fault monitoring, we propose one kind of dynamic network monitoring algorithm. This algorithm reduces the overhead caused by real-time network monitoring. By analyzing the probability of the alarm occurrence with the sampled network information, the proposed algorithm prolongs the interval between two consecutive pollings when no alarm could occur. This algorithm reduces the amount of monitoring traffic and guarantee to detect the abnormal event With the method of analyzing the performance of this algorithm theoretically we prove the validity of the given algorithm.2. In the aspect of network fault localization, we propose the MMDP network fault localization policy. The network fault monitoring system is a stochastic dynamic response system which means that the policy should also be dynamic with the network performance changes. The feasibility of the MDP has been proven theoretically. A policy for fault monitoring of a hierarchical network with the multi-manager is proposed by analyzing the multi-MDP. The middle managers can exchange the messages between each other to get the information of their sub-networks. Combined with the reinforcement learning, we get the optimized police in the limited period, thus causes the system reward maximization. We also determine the managed objects in the MIB as well as the polling structure, then point out the polling step increases along with the number of the managed objects increases and the polling step lengthens along with the observation stage reduces.3. In the aspect of the network fault diagnosis, we study the fast network fault diagnose with the application of the Bayesian Belief Network. The Bayesian Belief Network is an inference network according to the event probability. With the dependence relational model which conforms to the object-oriented network system, we prove the feasibility of the Noisy-Or and the Noisy-And network in BBN belief deducing. We also propose and prove the algorithm of the optimized method in finding the fault. The proposed algorithm has performances comparable to the one without the optimized method and significantly outperforms other existing fault finding algorithms.4. By analyzing the random ad hoc mobility model, we analyzing the application of the network fault management in ad hoc network and discuses the hierarchical architecture of the fault management Furthermore, the network fault management can be realized by GET and SET the managed objects whose state values are changed. The MIB information is given which provide a more complete application instance of network fault management.