Understanding Tradeoffs between Power Usage and Performance in a Virtualized Environment

Abstract

The usage rate of hardware resources in a computing system is the single largest influence on power consumption that can be impacted by software. Approaches to energy efficient scheduling on private clouds are not suitable for all situations, i.e. particularly when that approach is to consolidate VMs on as few machines as possible, a strategy that while reducing energy usage also impacts performance and dependability. In this paper, we introduce the concept of Workload Mixes, a technique for assigning tasks to compute nodes with the aim of balancing the trade-off between system properties such as energy consumption and application performance. In this paper workload mixes are demonstrated through a series of experiments to affect both system energy consumption and application performance in a virtualized environment. The exploitation of these effects may lead to an energy efficient, performance and dependability aware scheduler for a private cloud computing platform.