Research On Alarm Correlation Application In Communication Networks

With the extensive usage of communication services and rapid development of networkstandards, the scale and complexity of communication networks is growing rapidly. The typeand scale of fault alarm information is also becoming increasingly. To guarantee the reliabilityand stability of network operation, the alarm information of communication networks must bemanaged intelligently. And in alarm management, alarm storm has the most important effect.According to the characteristics of the networks and alarm storm, this thesis utilizescorrelation analysis in the alarm storm, and then focuses on the prevention, detection andprocessing of alarm storm. This thesis presents an improved FP-growth algorithm anddistributed weighted FP-growth algorithm for alarm storm prevention, and then designs thedetection alarm storm and the processing when alarm storm occurs suddenly.With analyzing the exiting FP-growth algorithm, the main problem of which is generatinga massive number of conditional FP-tree, the thesis presents an improved FP-growthalgorithm. The algorithm is processed by using AHP to determine the weights of alarminformation in different levels which makes mining more reliable. Then the algorithmintroduces a spare table to improve FP-tree. It avoids the generation of conditional pattern tree,with less affords in the mining process.According to the advantages of improved FP-growth and the distributed association rulefor data mining，the thesis presents a distributed weighted FP-growth mining algorithm. Thealgorithm consists of master node and slave nodes, using shared architecture. It avoidsgeneration of conditional pattern tree, and reduces the communication affords and thus makesthe mining more efficient when facing with massive alarm information.The thesis discusses the normalization of alarm from different manufacturers using theCORBA inform protocol, simplifying the interface of access. For the happening of alarmstorm, it adopts threshold means to detect. When the storm happens, it adopts rule strategy tofilt, remove and shield the alarm information.It describes the detection and the procedures ofalarm storm more comprehensively. To verify the performance and its effectiveness of theproposed algorithm, the thesis simulate the alarm information and topology of realcommunication networks, verify and compare with other algorithm.The results indicates thehigh efficiency and low affords of the algorithm presents in this thesis. It is also valuable anduseful for the prevention, detection and process of alarm storm in communication networks.