Dynamic consolidation methodology for optimizing the energy consumption in large virtualized service centers

In this paper we approach the high energy consumption problem of large virtualized service centers by proposing a dynamic server consolidation methodology for optimizing the service center IT computing resources usage. The consolidation methodology is based on logically structuring the service center servers hierarchical clusters, consolidation decisions being taken in each cluster using a reinforcement learning based algorithm. The methodology defines two ways of consolidation decisions propagation across the hierarchy: bottom-up propagation for the dynamic power management actions and top-down propagation for the consolidation actions. The consolidation decision time complexity analysis shows that the methodology usage in large service centers improves the decision time with a factor proportional with the ratio between the service center total number of servers and the logical clusters' number of servers.