Optimization of application placement towards a greener cloud infrastructure

Abstract

Cloud infrastructures are designed to simultaneously service many, diverse applications that consist of collections of Virtual Machines (VMs). The policy used to map applications onto physical servers (placement policy) has important effects in terms of application performance and resource efficiency. This paper proposes enhancing placement policies with network-aware optimizations trying to simultaneously improve application performance, resource efficiency and, as a consequence, power efficiency. The per-application placement decision is formulated as a bi-objective optimization problem (minimizing communication cost and minimizing the number of physical servers assigned to the application) whose solution is searched using an evolutionary algorithm with problemspecific crossover and mutation operators. Experiments carried out with a simulator demonstrate how a low-cost optimization technique results in improved placements that achieve all the target objectives.