Towards Accurate Accounting of Energy Consumption in Shared Virtualized Environments

Virtualized infrastructure providers demand new methods to increase the accuracy of the accounting models used to charge their customers. Future data centers will be composed of many-core systems that will host a large number of VMs each. While resource utilization accounting can be achieved with existing system tools, power metering is a complex task when per-VM granularity is the goal. In this paper we propose a novel methodology that brings new opportunities to power consumption accounting by adding an unprecedented degree of accuracy on the per-VM measurements. We present a system prototype that leverages power models based in performance monitoring counters (PMCs) to be used for energy accounting in virtualized systems. We validate the power modeling methodology in virtualized systems, by comparing the power predictions in both virtualized and non virtualized systems. The validation process has been performed as a case study for the Intel R © Core 2 Duo architecture. The validation follows two steps: first we validate the power model for one core, and second, we proceed on the validation of the entire processor. The resulting model is able to account for the power consumption for CPU and memory at process level. The main contribution of this paper is the introduction of a novel methodology that allows accurate accounting of energy consumption in virtualized systems. Accounting is done on a per-VM basis, even in the case that multiple VMs are deployed on top of the same physical hardware, overpassing the limitations of perserver aggregated power metering. Such approach can bring an unprecedented level of flexibility to the chargeback model used by service and infrastructure providers.