A Queueing Model Based Periodical ACPI Global State Control for Web Server Cluster Energy Management

Abstract

The queueing analysis of the call center staff planning is a common process for efficient resource allocation. In many ways, although at different time scales, this process is analogous to the provisioning of the nodes in a web hosting server cluster. Based on the well proven strategies learned from the call centers, this paper proposes an energy management method that (1) models a server cluster as an M/M/m queueing system, (2) periodically switches the ACPI (Advanced Configuration and Power Interface) global states of a server node, and (3) obtains an adequate number of active nodes constrained by a desired response time. The proposed process is simulated with a real world trace log. The simulation result shows that a significant energy saving can be achieved while maintaining a desirable response time.