Detecting Behavioral Variations in System Resources of Large Data Centers

The identification of significant changes in system resource behaviors is mandatory for an efficient management of data centers. As the dimension of modern data centers increases, the evaluation of state change detections through traditional algorithms becomes computationally intractable. We propose a novel approach that characterizes the statistical properties of the resource measures coming from system monitors, classifies them, and signals a change only when there is modification of the resource classification. This method diminishes the computational complexity and reaches the same detection accuracy of traditional approaches as demonstrated by several results obtained in real enterprise data centers.