Validating Heuristics for Virtual Machines Consolidation

Abstract

This paper examines two fundamental issues pertaining to virtual machines (VM) consolidation. Current virtualization management tools, both commercial and academic, enable multiple virtual machines to be consolidated into few servers so that other servers can be turned off, saving power. These tools determine effective strategies for VM placement with the help of clever optimization algorithms, relying on two inputs: a model of resource utilization vs performance tradeoff when multiple VMs are hosted together and estimates of resource requirements for each VM in terms of CPU, network and storage. This paper investigates the following key questions: What factors govern the performance model that drives VM placement, and how do competing resource demands in multiple dimensions affect VM consolidation? It establishes a few basic insights about these questions through a combination of experiments and empirical analysis. This experimental study points out potential pitfalls in the use of current VM management tools and identifies promising opportunities for more effective performance consolidation algorithms. In addition to providing valuable guidance to practitioners, we believe this paper will serve as a starting point for research into next-generation virtualization platforms and tools. 1