Towards VM power metering : A decision tree method and evaluations

In recent years, a large number of cloud data centers have been built around the world. It brings new challenges in the power management of data centers such as power monitoring, and scheduling for energy saving. All these challenges can be conquered much more easily if we know the power consumption of each virtual machine. Since VM runs at software level, modeling methods have been adopted to measure its power. However, current methods are not accurate enough, especially when multiple VMs are interacting with each other. In this paper, we propose a decision tree method to measure the power consumption of each VM. The advantage of our method is that the collected dataset can be partitioned into easy-modeling pieces by a best selected resource feature with proper value. We also propose a novel but simple method to evaluate the accuracy in a more objective way. We use standard deviation of errors to evaluate the stability of our method. Experiments show that our method can measure the power consumption of VM with high accuracy and stability.