Deep Reinforcement Learning Based DVFS Algorithm Frameworks

Energy consumption is a key problem in modern society. Computing devices, e.g., personal computer, smart watches, data centers are widely applied. Therefore, how to reduce energy consumption in some specific scenarios becomes a crucial research question. We study Dynamic Voltage and Frequency Scaling (DVFS) problem in periodical task scenario. We first analyze two classic algorithms step by step. Then, we propose our new solutions which consumes less energy in the same task set compared to classic algorithms. Furthermore, we use deep learning techniques in our learning based DVFS frameworks. The new algorithms and frameworks contribute to energy-saving society.