Energy Efficient and Interference Aware Provisioning in Virtualized Server Cluster Environment

Abstract

IT service providers, employ server virtualization as a main building block to improve system utilization, improve system manageability, reduce operational costs which includes energy consumption driving economies of scale with shared resources. Virtualization enables co-locating and efficient assignments of virtual servers within the bounds of limited number of heterogeneous physical servers, with Virtual Machines (VM) sharing the limited physical server resources between them. Though virtualization technologies point to the fact that each virtual server has its very own isolated environment, but in reality, perfect isolation is not possible. Primary measure to achieve assignment efficiency is to ensure that system resources are utilized effectively and performance of VM (and application workloads) is consistent within the desired bounds. Interference or contention on the limited shared resources among VMs leads to performance degradation and is referred to as performance interference. This affects (a) application Quality of Service (QOS) and (b) server cluster or data centers' energy-efficiency. In this work, we analyze the performance degradation using (a) energy efficiency heterogeneity measure and (b) interference aware measure, with the aim to reduce energy consumption in our environment. Experimental results on different scenarios with our energy efficiency and interference aware approach shows a reduction in energy consumption to the tune of 8 to 58% and 10× improvement in per request average response time in contrast to a default energy efficiency and interference oblivious approach.

Power management using dynamic power state transitions and dynamic voltage frequency scaling controls in virtualized server clusters

“…We briefly discuss different models to arrive at server heterogeneity factor PPM, power, energy, and performance; system architecture; and an algorithmic approach using these defined models. In this paper, we have extended model notations and simulation test-bed setups from our earlier work [14] to account for DVFS and DPM sleep state transitions.…”