Dynamic configuration of virtual machine for power-proportional resource provisioning

Power management is a major concern in datacenters. Using virtualization in datacenters enables applications' consolidation to reduce power consumption. However, processors consume the most fraction of the host's power. Furthermore, the rising number of cores in a single processor extensively contributes to the increase of power consumption if there are no efficient power management solutions. These solutions are considered inefficient if they do not take into account the number of active physical cores and the configuration of a virtual machine, which runs a certain job. In this paper, we analyze power consumption of a multicore processor and develop a CPU power model and a performance model based on the number of active cores and frequency. Then, we propose an optimization solution for power and performance management in virtualized servers. Our optimization model achieves power proportionality and guarantees performance; it is based on a mixed integer programming model. The optimization model provides an optimum configuration for both a host and its VMs in terms of their number of virtual CPU and their proportional weight. Finally, we demonstrate efficiency of the proposed solution via experiments. The results show that between 23% and 48% savings in power consumption compared to a typically provisioned power by hypervisor performance governor.