A Nonparametric Multivariate Method for Performance Analysis of Virtual Machines in Cloud Computing Systems

Chang Liu, Ming Yu, and Yu Zhang


Load modeling and forecasting, performance analysis, cloud computing, nonparametric multivariate analysis


Performance analysis of virtual machines is an indispensable capability for achieving on-demand resource provisioning in cloud computing systems. In this paper, a nonparametric multivariate method is presented as a solution to forecast performance degradation of virtual machines in cloud computing systems. Firstly, the k-means algorithm is adopted to partition multivariate training data into three clusters, which correspond to the virtual machine states of normal, anomaly and failure. Based on this clustering, the performance data of working virtual machines are classified. For those classified as anomalies, a nonparametric CUSUM algorithm is carried out to analyze whether they will lead to serious performance degradation (corresponding to the failure state). Experiment results based on Hadoop show this method can not only identify the normal and failure states of virtual machines, but also succeed in forecasting performance degradation of virtual machines by those anomalous data.

Important Links:

Go Back