Performance Prediction of Network-Intensive Systems in Cloud Environment: A Bayesian Approach

The resource requirements of cloud applications varies with the load intensity, so it is necessary to predict resource changes in advance. In this paper, a Bayesian method is proposed to predict the bandwidth consumption and quality of service. First, log mining is used to extract feature variables, and a Bayesian model is constructed by using stepwise discriminant method to select characteristic variables that are crucial for prediction. Then Bayesian discriminant and Bayesian network are used to predict the bandwidth consumption and response time. At last, we use a real-running web system to evaluate the effectiveness and accuracy of this method. Comparing with other methods, it can get more accurate prediction result of the performance in the cloud environment.