Abstract: Aiming at the impact of Dynamic Voltage Frequency Scaling(DVFS) on program’s performance and power consumption,based on interval division,this paper uses existing performance monitoring unit provided by commercial processor,it proposes a performance prediction model which takes the relation between memory access latency and frequency into account. It presents an energy-optimized DVFS(eDVFS) control policy that aims to minimize energy consumption. Experimental results show that eDVFS can get 23% at most and 6.85% on average in energy saving compared with Linux ‘ondemand’ adjusting policy.

Key words: Dynamic Voltage and Frequency Scaling(DVFS), power consumption prediction model, performance prediction model, voltage and frequency scaling mechanism, energy consumption optimization

摘要： 针对动态电压频率调节(DVFS)对应用程序运行时性能与功耗的影响,基于区间划分方法,使用现有商用处理器提供的性能监测单元,提出一种考虑访存延迟变化的DVFS性能预测模型,并利用该模型实现针对能耗优化的DVFS调节机制(eDVFS)。实验结果表明,与Linux内核提供的ondemand调节策略相比,该eDVFS调节机制能够获得最大23%、平均6.85%的能耗优化。

关键词: 动态电压频率调节, 功耗预测模型, 性能预测模型, 电压频率调节机制, 能耗优化

CLC Number: 

• TP391