A Novel Server Consolidation Method Based on Local Storage Integrated with Resource Demand Prediction

Server consolidation plays a significant part in energy-saving technology in data centers. Traditionally, cloud service instances commonly use shared storage architecture. Nowadays, data and I/O intensive applications are preferred in this big data era and are used in the majority of Internet companies, much more attention has been paid to the local storage that offer perform better in I/O at a lower price compared with shared storage clouds. But these cloud instances usually contain much more data than shared storage cloud instances. Thus, in such local storage based clouds, the migration cost can be really high. Unfortunately, most existing work about did not consider integrating the demand prediction algorithm that plays a significant part in server consolidation, especially for local storage based cloud, where the migration cost is very high and is in badly need of an efficient resource pre-allocation mechanism. To address this issue, we proposes Aricon, a consolidation method based on local storage. Our approach uses a time series model to forecast the CPU or memory utilization of instances within servers. We investigate the effectiveness of instance and server resource utilization prediction in server consolidation performance in workload traces from real world. To validate the performance of the proposed Aricon, we test the prediction accuracy and compare it several existing consolidation method, and the results show that Aricon not only has low prediction error rate in 10.7% but also schedules computing resources efficiently.