Multi-Class Queuing Networks Models for Energy Optimization

Abstract

The increase of energy consumption and the related costs in large data centers has stimulated new researches on techniques to optimize the power consumption of the servers. In this paper we focus on systems that should process a peak workload consisting of different classes of applications. The objective is to implement a policy of load control which allows an efficient use of the power deployed to the resources. The proposed strategy controls the workload mix in order to achieve the maximum utilization of all the resources allocated. As a consequence, the power provision will be fully utilized and the throughput maximized. Thus, the costs to execute a given workload will be minimized, together with its energy consumption, since the required processing time is decreased.