An Energy-Efficient Management Mechanism for Large-Scale Server Clusters

With the increase of the computing demand, high performance server clusters are becoming one of the most important computing infrastructures. The current clusters are designed to meet peak load with all the computing resources keeping running. However, this static reservation with full computing resources can not adapt to the time-varying computing requirement, and may incur low resource utilization and needless power consumption when the cluster system is underloaded. In this paper, we present an extensible architecture of cluster management system. This architecture promises a good extensibility by integrating job scheduler and resource manager in loose couple. Concentrating on the power saving of large-scale clusters, we describe the power model of servers, and based on the presented management system architecture, we propose a novel resource management way, adaptive pool based resource management (APRM) method, for adaptive provision of computing resources in accordance with the time- varying workload demand. APRM enables a cost- effective operating by providing dynamic computing capacity with automatic resource control. We validated APRM on the energy efficiency and quality of service (QoS) by simulation measurement, and the results showed that APRM yields significant power saving with little impact on QoS.