A Prediction Based Server Cluster Capacity Planning Strategy

摘要

Cloud computing is an Internet-based service which provides shared virtual resource and data to accomplish certain computation. In order for the servers to have sufficient resources when the request arrives, as well as save server resources as much as possible, we propose a prediction-based server capacity planning and dynamic scheduling algorithm. There are mainly three steps in our capacity planning algorithm. The first step characterizes the given data on several indices and then present an effective model in order to predict the oncoming demands in the near future. The second step generates the workload of servers combined with the predicted demands and then make capacity planning based on this workload. Thus it's obvious that the effectiveness of capacity planning depends on the accuracy of prediction to a great extent. Finally, a demand prediction based strategy on workload allocation is brought out. A dynamic resource allocation strategy is given to ensure the quality of service at any moment in future meanwhile taking energy consumption into consideration. The results of the experiment show that the required server number decreases by 33% after the prediction-based capacity planning applying on server scheduling.