A self-adaptive DVS scheduling algorithm for multi-core systems

Abstract

Dynamic Voltage Scaling (DVS) is widely supported in modern general purpose systems, like desktops. Interval based DVS algorithms have been used in some operating systems (OSs), like Linux. But their load prediction models can't fit various needs of users so well, since most of them employ one prediction method only. In this paper, a self adaptive algorithm called SADVS is presented. It selects the best algorithm according to the prediction accuracy of historical records, and the selector will be used to adjust the CPU speed in the following interval. SADVS behaves well in both simulations and actual implementation. For actual implementation, we find that SADVS reduces energy consumption 0.02% .... .. . 4% without time extension compared to results of ondemand policy in Linux kernel.