Research And Implementation Of Fault Localization Algorithm For IP-Based Networks Using Bayesian Networks

As a significant function of network management, fault localization has played an increasingly important role in the IP-based networks to guarantee the availability of the network and the continuous of the enterprise service. In particular, with the rapid development of IP-based networks, various enterprise services and the continuous improved demand for service quality, how to locate the root cause of the fault more quickly and more precisely so as to solve the problem when faults appear is sure to be the key problem of the network management system. Currently, user's demands for service quality have improved a lot, and high-quality service depends on the fast and accurate response of the fault, so fault localization is sure to be an increasingly important role. Accompanied by the expansion of the network and the enterprise service, IP-based networks shows some new characteristics such as uncertainty, complication and so on, which mount new challenges to the rapid and accurate fault localization. Therefore how to do effective fault location in a complex, large-scale, noise and uncertain network will be the focus of the research of fault localization algorithm.Bayesian networks can model both deterministic and probabilistic dependencies among many types of the faults and the symptoms, so that it can accurately response the various relationships between the entities in the IP-based networks. Furthermore, the Bayesian networks uses the graph-structure to express the joint probabilistic distribution of variables and conditions of independence, which can greatly simplify the inference complexity even in large-scale IP network. So Bayesian networks is especially suitable for reasoning about failures in the IP-based networks which has a lot of enterprise service.Based on an application scenarios, this paper proposes a new fault localization algorithm based on bayesian networks by the research of new needings of fault localization in the IP-based networks and doing amend to the exist algorithm. This algorithm introduces the limit number of simultaneous faults and does an approximate inference so that it can speed up the fault localization process, and at the same time it provides nice localization accuracy. Because this algorithm can simultaneously use both positive and negative symptoms and introduce a technique called analysis threshold to filter spurious symptoms, it has nice noise resisting ability. At last in the experiments we compare our algorithm with a prior algorithm IHU( We also implement this algorithm with java to process the experiment.) . Results show that our algorithm gets similar accuracy as IHU but with much less localization time, our algorithm performs better in realistic noisy networks. It could be seen that our method is more suitable to be used in the large-scale IP-based complex networks.