Risk management for virtual machines consolidation in data centers

Virtual machines (VMs) consolidation has emerged as an important method for the design of energy-efficient data centers. The purpose is to aggregate VMs to fewer physical machines and put the idle servers into power-saving mode. Existing researches mainly focus on transforming the VMs consolidation into various bin packing problems. However, VMs consolidation may cost Service Level Agreement (SLA) violations just after the migration due to the uncertainty of applications' demands. In this paper, we provide a SLA risk management framework, involving a stochastic program to solve the resource allocation for VMs and an algorithm for dynamic VMs consolidation at runtime, to optimize both the energy consumption saving and SLA violations. We validate the proposed algorithm using workloads from a real world system. The results compare with other VMs consolidation algorithms that without considering risk, and show that our SLA violations is reduced by four times from 25% to 2% - 5% while only losing little energy consumption saving.