Adaptive Dynamic Power Management Policy: Two Value Alternate Basis

Abstract

In dynamic power management, fixed time-out policy is not optimal because of waiting before shutting down a component and adaptive time-out policy is instability for re-choosing parameters for different workloads. To handle these problems, a new adaptive time-out policy is proposed in this study. Combining the advantage of fixed and adaptive time-out policy zero and spin-down cost are chosen as time-out values in the policy. They alternate as time-out values according history workloads. Zero as one of time-out makes for saving more power. And spin-down cost which only depends on disk model causes policy has more steady performance. Policy based on tow values sufficiently uses the character of cluster and bursty of workloads. Simulation results show that policy proposed in this study is approach to optimal policy and performs more steadily.