Finding relevant dimensions in Application Service Management control: A features selection approach

In recent years there seems to be an increased interest in autonomous control in Application Service Management environments. This is effectively causing fast growing demand on analysis of multivariate datasets in the area. Specifying a causal model in the controlled system significantly simplifies evaluation of defined elements utilization dependencies. This allows more efficient search for similarities in the time-series, selection of most relevant dimensions, and easier control in the reduced space, which would ultimately reduce maintenance effort. This paper proposes the feature selection method based on metrics time series analysis. The proposed method performs multivariate evaluation tackling the strength of dependency search of metrics sequences from three different perspectives: Similarity, Consequence, and Interference; all these factors are then jointly considered in the calculation of Clarity, which is the final dependency measure of the proposed algorithm (SCIC). Using SCIC, we show that the technique can be applied in the service control practice and evaluated from engineering perspectives.