Accuracy-Enhanced Power Metering Technique in Virtualized Environments

Abstract

In virtualized environments, accurate metering the power consumption of individual virtual machine (VM) is a challenging issue. Conventional VM power metering techniques rely on the assumption that the power consumption is linear to the utilization of hardware utilization. However, such a utilization-based technique can only provide coarse-grained power measurement with unbounded error. In this paper, we firstly formulize the relationship between the resource utilization and the accuracy of power metering. Then, we proposed a novel VM scheduling algorithm, which uses the information of performance monitoring counters (PMC) to compensate the recursive power consumption. Theoretical analysis indicates that the proposed algorithm can provide bounded error when metering per-VM power consumption. Massive experiments are conducted by using various benchmarks on different platforms, and the results shown the error of per-VM power metering can be limited below 5.2%.