A frequency-aware management strategy for virtual machines in DVFS-enabled clouds

Abstract

It is demanding to curtail energy consumption of virtual-machine-powered data centers, because modern data centers have been significantly scaling up in capacity in past decades. In this study, we propose a frequency-aware management strategy, which controls dynamic power and static power of processors running virtual machines in data centers. Unlike existing dynamic voltage and frequency scaling schemes, our strategy simply incorporates frequency requirements rather than task execution times. This salient feature is practical, because task execution times in a raft of real-world applications are unknown in a priori. We build a frequency-aware model, which is adept at deriving an optimal frequency ratio that minimizes processors’ energy consumption. With our model in place, the energy efficiency of a data center can be maximized by adjusting the processor's frequency to meet the optimal frequency ratio. We design a management approach to judiciously adjust frequency ratio to conserve energy without violating the frequency requirements imposed by virtual machines. After analyzing the correlations between frequency ratio and energy consumption, we show that a small static power proportion gives rise to high energy-saving performance. The results demonstrate that our model lays out a solid theoretical foundation catering to the development of power management software in DVFS-enabled clouds.