The power-saving opportunities of long-running application servers which execute on multi-core systems are studied in this paper. The research goal is to develop an efficient power-saving strategy of application servers with the minimum performance degradation in cloud environments. The power-saving strategy is based on the run-time information which is already available in a JVM, the base software component of application servers. Several key findings are revealed through this study. First, the particular behavior of application servers, also known as phases, can be related to the run-time information of a JVM. Thus the phases of an application server can be predicted before the applications actually execute on hardware. Secondly, some particular phases are observed in this study and used to establish the power-saving strategy, such as memory phases and execute phases. Finally, a new finding of idle phase is proposed to reduce significant energy wastage without performance degradations. Based on these findings, a set of power-saving algorithms is proposed and implemented with two widely used JVMs, Sun's Hotspot and Jikes RVM. With the experiments of five multi-threaded benchmarks and two web application benchmarks, the use of proposed power-saving strategy leads to the lowest value of energy-delay product among the other power-saving techniques, and the performance degradation is well below 6 percent.
All Science Journal Classification (ASJC) codes
- Information Systems
- Hardware and Architecture
- Computer Science Applications
- Computer Networks and Communications