基于随机决策模型的动态功耗管理策略研究  

Study on Policy of Dynamic Power Management Based on Stochastic Decision Models

作　　者：吴琦[1] 熊光泽[2] 

机构地区：[1]中国科学院空间科学与应用研究中心,北京100080 [2]电子科技大学计算机科学与工程学院,成都610054

出　　处：《计算机学报》2007年第4期622-628,共7页Chinese Journal of Computers

基　　金：国家"八六三"高技术研究发展计划项目基金(2003AA1Z2210)资助

摘　　要：由于功耗的严格约束,现代嵌入式计算终端必须采用科学的动态功耗管理策略.文中在对计算机系统的动态功耗管理(Dynamic Power Management,DPM)模型深入研究的基础上,采用改进的DPM随机决策模型,从理论上证明了DPM最优策略是确定性马尔可夫策略,这为简化DPM控制算法提供了理论依据.在实例研究中,比较了空闲时间长度服从负指数分布与Pareto分布两种情况,发现经典的空闲时间长度服从负指数分布的假设与实际情况偏差很大.Pareto分布很好解释DPM超时策略在实际应用中可以取得优良节能效果这一现象.Because of the energy consumption limitation of modern embedded computing terminal, the reasonable policy for the dynamic power management （DPM） must be used. Based on the deeply study of DPM model for the computer system it is proved theoretically with stochastic models that the DPM optimal strategy is a deterministic Markov control strategy. The conclusion provides the theoretical basis for the simplification of DPM algorithm. In example, two idle time distributions are compared. The one is the negative exponential distribution and the others is two parameters Pareto distribution. It is found that the exponential distribution supposition taken by traditional queuing theory is not suitable to DPM. The practical effect of the time-out strategy is well interpreted with the Pareto distribution.

关 键 词：动态功耗管理 PARETO分布 马尔可夫决策过程 

分 类 号：TP316[自动化与计算机技术—计算机软件与理论]