CPU Usage Prediction Models for Virtualized Data Center

Abstract

Resource allocation plays an important role in Virtualized Data Center (VDC). The applications running in VDC are mostly business critical applications with Quality-of-Service (QoS) requirements. Moreover, dynamic resource allocation and real time monitoring of the resource usage of VMs are also needed to reduce under resource utilization and over resource utilization. Therefore, resource usage prediction is required for dynamic resource allocation systems. In efficient dynamic resource allocation, the resources are allocated to a VM while meeting their Service Level Agreement (SLA). The main contribution of this work is two-fold. The first is the generation of CPU usage prediction models by applying different powerful machine learning techniques. The second is SLA evaluation on predicted value by using proposed SLA metric. To evaluate the efficiency of these models, experiments are carried out by using CPU profiles from real world data centre. According to the experiments, proposed resource prediction models have promising accuracy.