A memory-aware energy saving algorithm with performance consideration for battery-enabled embedded systems

Dynamic voltage and frequency scaling is an important mechanism for reducing the energy consumption of a processor while tasks are running. Our studies indicated that the lowest energy consumption usually appears at some frequency other than the lowest frequency. In addition, there exists an inverse relationship between memory access rate and the frequency which can minimize the energy consumption. A correlation equation can thus be deduced from the relationship and used at task execution time to find a frequency which tends to minimize the energy consumption. In this paper, an energy saving DVFS algorithm based on the correlation equation is proposed. It maintains the performance at a reasonable level. The algorithm has been implemented on Linux as a user-space power manager. The experiment results show that our algorithm performed better than the commonly used Linux Ondemand DVFS algorithm, according to the energy-delay production metric.