Failure Prediction based on multi-parameter analysis in support of autonomic networks

In this paper, we present a Failure Prediction System (FPS) using a novel algorithm that extracts frequent anomalous behaviors based on multi-scale trend analysis of multiple network parameters. The proposed Correlation Analysis Across Parameters algorithm (CAAP) utilizes multiple levels of timescale analysis to reveal the frequent anomalous behaviors. The CAAP philosophy is that failures usually do not occur because of change in a single parameter behavior; instead, a set of interrelated parameters change their behaviors jointly and lead to a particular failure. The proposed algorithm requires an enhanced version of FABM algorithm which was presented by the authors in a previous paper and was used to analyze each parameter's behavior individually. Moreover, the new version, called FABMG algorithm, has the same polynomial computational complexity of O(n²). The CAAP utilizes the data mining techniques of association rules mining in order to reveal the existed correlation relationships. Consequently, as found in this work, this approach improves the quality of the FPS results which was relying on individual parameter analysis only. One of the strengths of CAAP is that it requires the FABMG output only, i.e. it does not require rescanning the database in order to produce the correlation results.