Computing System Performance Anomaly Detection Algorithm To Achieve

The computing system performance exception is the phenomenon of performance characteristics smoothly degrading during the running of software, because of exhausting of resources or cumulating of runtime errors, until an intolerable reduction of capacity is eventually reached. The performance anomaly detection can quantificationally analyze the performance attributes according to the engrossing of system resources and real-timely inspect system's status.Firstly, we investigate the algorithms of performance anomaly detection. They are anomaly detection algorithm bases on statistic, anomaly detection algorithm bases on data mining, anomaly detection algorithm bases on machine learning and anomaly detection algorithm bases on natural immunity system. We also analyze the shortage of existing algorithms. They can not completely satisfy the need of performance exception detection. Secondly, we aim at the special problem of performance exception detection and adopt general anomaly detection algorithm, in which healthy system states are divided into one class, unhealthy states are divided into another class. Thereby the multi-phased clustering algorithm is proposed, it integrates layered clustering and other clustering to form multi-phased clustering. Thus the algorithm can effectively increase the accuracy and the rate. The experimental results show that the proposed algorithm has a good detection effect. Thirdly, the improved Bayesian classifiers is proposed, naive Bayesian classifiers which make independence assumptions perform remarkably well on some data sets but poorly on others. We explore ways to improve the Bayesian classifier by searching for dependencies among attributes. The improved classifier boosts its applicability and increases the classified rate. The experimental results show that the improved classifier can effectively improve the classified accuracy and the detection effect. Finally, we compare the two algorithms.