A Big Data Processing-oriented Prediction Method of Cloud Computing Service Request

In order to guarantee the cloud service quality, the service should be able to dynamically predict the change of data processing request. Existing prediction methods in cloud are mostly focused on the amount of computing resource required by service. In fact, in cloud computing environment for big data processing, it is not enough to simply predict the computing resource, because when the created virtual machine is far from the data, it will need a certain time to transfer data to the virtual machine for processing. To solve this problem, in this paper, we propose a data-centered prediction method using Bayes classifier, which can make prediction for data type or location based on the data resources needed by the service request. We carry out experiments with Google cluster trace, and the experimental results show that our method performs better than the existing methods. For example, our method improves the load prediction accuracy by 45-60% compared to other state-of-the-art methods based on final state-based method, simple moving average method, linear weighted moving average method, exponential moving average method, and prior probability-based method.