Fault Management Based On Event Correlation And Data Mining

In this information age, the computer network plays a more and more important role in people's production activities and daily life. Once a fault occurred in the network, network engineers and network managers should make quick and accurate judgments about the type and the source of the fault in order to make a plan for recover. A lot of research on network fault management is conducted in this paper, basing on event correlation and data mining technology.The conception of rule element is introduced into traditional rule-based network fault management system to share the same rule element between different rules and compress the rule base. At the same time it also facilitates rule elements sorting, group creation, index creation and common rule elements buffering. All this can improve the speed and efficiency of rule element query. Besides that, rule element is used instead of rule to match with facts. This reduces the cost of matching and the occurrence of useless matching. Therefore it improves the efficiency of Inference Engine.A new association rule mining algorithm is proposed basing on address index matrix. It goes through the alarm transaction database once to convert it into matrix and save the address of frequent item. Through this, the speed and efficiency of query is improved.A new sequence rule mining algorithm is proposed basing on sequence matrix SM. All kinds of alarm data are taken into account in this algorithm. It mines the sequential pattern by setting different time windows. Besides that, the introduction of topological constraint for effective pruning also speeds up sequential pattern mining.By applying event correlation and data mining to network fault management, we design a complete rule inference based network fault management system model, which support both association rule and sequence rule for inference. In addition, a detailed design is made. At last, the rule-based reasoning module is developed, and it proved the effectiveness of association rule and sequence rule reason.