Integrated Network Operation Baselining and Adaptive Detection of Faults and Performance Problems(Network Quality Control)(Special Issue on Collaboration Art and Network Entertainment)

概要

論文の詳細を見る
Detection of network problems is a crucial step in automating network management. It has a direct impact on the accuracy of fault, performance and security management functions. In this paper, we propose an adaptive technique for network fault and performance detection. Network operations normal baseline is parameterized as a finite mixture model, and model parameters are estimated from data using the Expectation Maximization algorithm. We propose a novel method for online residual generation, based on repeated identification of model parameters. The resulting multivariate residuals are shown to be approximately Normal with the mean 0 and unit variance under normal conditions, and the mean jumps suddenly to a different value, once anomaly occurs. To achieve fast detection, we formulate the problem of detection as a sequential disruption problem, where the task is to detect departures from the baseline distribution as soon as the disruption happens. An analytical expression of false alarm rate allows us to choose the threshold, automatically. Experimental results on a real network showed that the monitoring agent is able to detect even slight changes in the characteristics of the network, while maintaining a low false alarm rate.
一般社団法人情報処理学会の論文
2003-02-15