A Power State Transition Algorithm for Power Controller of Energy Efficient Computer Clusters

概要

論文の詳細を見る
With the increase in both performance and scale of computer clusters, energy consumption has attracted much attention because of their impact on the environment and inevitably increasing operational cost. While various different approaches are being proposed to reduce power consumption of clusters, one promising approach is to make the idle nodes into low-power consuming state ( i.e., turn-off or sleep) when the system's utilization level is low. However, complex dependencies among physical resources and services make it difficult to create common power control tool for such clusters. In this paper we propose an algorithm which solves complex dependencies among cluster nodes including both physical and service nodes. In addition, since different facilities offer various kinds of services and each has unique structure, they have different system management policies and constraints. This situation makes it hard to create a common power control tool that works on different kinds of clusters. Therefore, by exploiting our algorithm, we propose a versatile, scalable power control tool, ClusterNap, with which users can define their own policies and constraints that explicitly suits for the physical resources they have and the services they provide.
2013-07-24