Predictive Control for Energy-Aware Consolidation in Cloud Datacenters

Abstract

Infrastructure-as-a-Service is one of the most used paradigms of cloud computing and relies on large-scale datacenters with thousands of nodes. As a consequence of this success, the energetic demand of the infrastructure may lead to relevant economical costs and environmental footprint. Thus, the search for power optimization is of primary importance. In this perspective, this paper introduces an energy-aware consolidation strategy based on predictive control, in which virtual machines are properly migrated among physical machines to reduce the amount of active units. To this aim, a discrete-time dynamic model and suitable constraints are introduced to describe the cloud. The migration strategies are obtained by solving finite-horizon optimal control problems involving integer variables. The proposed method allows one to trade among power savings and violations of the service level agreement. To prove its effectiveness, a simulation campaign is conducted in different scenarios using both synthetic and real workloads, also by performing a comparison with three heuristics selected from the reference literature.