Cloud Applications Consolidation through Context Information and Heuristic Optimization

Resource consolidation has been proposed as an effective mechanism to save energy in data centers. Several algorithms have been developed for this purpose, which use the minimal number of servers and network switches, and power off unused equipment. Consolidation algorithms usually take into account static service constraints (Central Processing Unit CPU, Random Access Memory RAM, disk, bandwidth), but do not consider dynamic context information, as CPU utilization and the specific role of each Virtual Machines (VMs) in the running application (e.g., core component, backup replica, member of a pool of workers for load balancing).In this paper, we describe and evaluate a novel heuristic based consolidation strategy that explicitly considers context information. Our approach avoids running idle VMs that are pre-provisioned for availability and redundancy purposes, hence pursuing a better linear relationship between power consumption and actual computation than other existing algorithms. We demonstrate through simulations, by comparing different heuristics, the optimal trade-off between service level and energy efficiency achieved by our approach.