Research And Implementation Of Power Management System For High Performance Computer

With the rapid development of high performance microprocessors, high-speed networks, and standard middleware tools, supercomputers have been widely used to provide amazing computing capability for both industrial and scientific applications. However, huge power consumption has prevented the further application of the large-scale computer systems. Reducing the power consumption is now a critical issue in designing high performance computers. Based on a deep investigation and analysis of the current research and related works in this field, this thesis presents a prototype system which aims at reducing the energy consumption in running the high performance computer system. Design and implementation approaches are described in detail.First, we propose an online power-saving state control model and an energy consumption model for heterogeneous clusters, then a novel energy-aware task scheduling algorithm for heterogeneous clusters is proposed, which is based on the general adaptive scheduling heuristics Min-Min algorithm. The algorithm is evaluated on a simulated heterogeneous cluster. The experiment results show that the new energy-aware algorithm can achieve a good time-energy trade-off and outperform the original Min-Min algorithm under various conditions.Based on the concept of the above algorithm, we propose a novel mechanism to reduce energy consumption at the system level of high performance computers, and implement a prototype system according to it. Software architecture, module decompsition and detailed implementation strategies of the system are also presented.According to the requirements of our project, a test plan was developed. Our prototype system was evaluated using real workload under real environment. Analysis to the results shows that our method and mechanisms are both practical and effective.