Title :
An adaptive power management with policy selection mechanism based on reward-punishment scheme
Author :
Fa-Gui Liu ; Xiao-Yong Xing ; Jin-Biao Lin ; Jun Lin
Author_Institution :
Sch. of Comput. Sci. & Eng., South China Univ. of Technol., Guangzhou, China
Abstract :
Dynamic Power Management (DPM) is a technique to reduce power consumption of electronic system by selectively shutting down idle components. The control algorithm (DPM policy) is the core in DPM system. Different policies have their own best performance in some specific system running environment. In the other hand, system running environment changes with time. Most policies cannot adapt to changes in system running environment.. This paper proposes a policy selection mechanism based on reward-punishment scheme, which can dynamically select a proper policy from a set of policies according to the system status and the past performance of policies at every decision time. Experiments prove that the proposed algorithm can achieve better performance than any other single policy synthetically.
Keywords :
computer networks; low-power electronics; power consumption; power control; telecommunication control; telecommunication power management; DPM policy; DPM system; adaptive power management; control algorithm; dynamic power management; electronic system; policy selection mechanism; power consumption; reward-punishment scheme; Abstracts; Adaptation models; Observers; Servers; Software; Adaptive Power Management; Policy selection; Reward Punishment Scheme;
Conference_Titel :
Machine Learning and Cybernetics (ICMLC), 2013 International Conference on
Conference_Location :
Tianjin
DOI :
10.1109/ICMLC.2013.6890878