一种自适应的动态功耗管理预测策略

An adaptive dynamic power management predictive policy

动态功耗管理通过关闭空闲的系统部件来减小电子系统的功耗。如何提高预测的准确度是当前基于预测动态功耗管理(DPM)研究的主要问题。基于预测的DPM策略假设系统部件访问时间具有关联性,当前应用最为广泛的是指数滑动平均方法,但其应对突发事件的能力不强。本文基于指数滑动平均改进模型,对突发事件进行数学定义,提出一种自适应的DPM预测策略,在面对突发的长延时或短延时情况下,系统都能自适应调整,提高了预测的准确度,优于现有策略。

Dynamic Power Management（DPM） is a design methodology aiming at reducing power consumption of electronic systems by performing selective shutdown of idle system resources.How to improve predictive accuracy is the main problem needed to be solved in DPM policy based prediction.With the assumption of the relevancy among the pieces of idle time,the exponential average policy Under is prevalently used can not predict accurately when abnormal error like a sudden long or short idle period happens.In this paper,the authors propose a new adaptive DPM predictive policy basing on improved exponential average model and mathematic definition of abnormal error.The policy would self-adjust to get higher predictive accuracy if sudden error occurs.The new policy excels the existed ones in prediction accuracy.