Adaptive Workload Prediction for Cloud-Based Server Infrastructures

Abstract

Currently, data centers offer cloud computing platforms relying on virtualization technology and multi-tier architecture to handle an ever increasing scale and to provide elastic service. However, in order to achieve elasticity, efficient prediction is needed to control virtual machines. We present a new adaptive linear auto regressive technique for web server workload prediction with feedback loop control. We test the Adaptive-Feedback AR model with the Songkhla Rajabhat University’s academic web which has a similar daily pattern of workloads and the learning management system (LMS) web which has unpredictable workloads. For the 1-minute interval, the suitable result for controlling the AR orders is in the range of 2-8 and previous historical value is in range of 10-25. Our new prediction approach predicts both web workloads with a root mean square error (RMSE) below 0.6, of which quality is better, in terms of the prediction accuracy resulting in a better performance.