Research And Implementation Of Forecasting Method In Cloud Resource Management

With the development of cloud computing, virtual resource management has become one of the key technologies. In order to improve resource utilization to obtain greater benefits, cloud computing requires effective management on resource. Through the analysis of some virtual resource management methods, we found that these methods adopt prediction technique. The workload of application is usually dynamic change and the resource it needs changes too. Because there is a certain lag time of launching resource, we should prepare it in advance. To solve this problem, the prediction technique is introduced into the cloud resource management.Analyzing workload, we found that in workload there is a certain pattern: cycle mode, non-cycle mode or hybrid mode. At present, there is no prediction model which can be suitable for all modes. In order to predict workload more accurately, we design pattern sensitive prediction method. This method can identify the pattern of workload and select an appropriate prediction model to predict. Meanwhile, this method can sense the changes in the workload pattern and adjust prediction model adaptively to respond these changes. In prediction error is inevitable, to reduce the prediction error, we use artificial neural network (ANN) and Markov chain to correct the predicted value.In order to verify the accuracy of the prediction models and the adaptability of the prediction method, we use virtual machine’s CPU utilization as the experimental data. Experimental results show that our prediction models can conduct good prediction on the workload in corresponding pattern; our prediction method can also be adaptively sense the change in workload pattern and adjust the prediction model; our error correction method can improve the prediction accuracy further.