Design And Implementation On Power Management Techniques For High-performance Computing System

The field of High-performance computing has experienced extensive changes andremarkable development. High-performance computing has been widely used by manycompute-intensive applications involving energy, biology, meteorology, Geology andscientific research. High-performance computer is an outstanding symbol ofComprehensive national strength in the field of science and technology. It is also playan important role in modern society. On the other hand, the power consumptions ofhigh-performance computing system are already tremendous. It poses a seriouschallenge to the development of high performance systems.With the continued development of semiconductor technology and significantlyimprovement of chip integration, microprocessor’s performance and power density areboth increasing sharply. At the same time, the scale of these systems becomes larger andlarger, and the density of chipsets integration increases drastically because of the greaterdemands on performance. High power consumption will lead to high heat and consumemore cooling costs, and greatly limiting the development and application ofhigh-performance computers. In this paper, we use Simshedule(a parallel system jobschedule simulator) to analysis the characteristics of workload and power-usage ofcomputing nodes and different component in high-performance parallel computingsystem. Based on this analysis, we propose a power management system seek to avoidthe loss of system performance while reducing overall system power consumption.A power-aware algorithm based on parallel resource management was proposedwhich can optimizes the energy efficiency of system by control the node’s threshold ofidle time period. A node will be switched into “sleep” mode when its idle time periodexceeds the threshold. Detailed experimentation using traces from the ParallelWorkloads Archive indicates that there algorithms can achieve considerable overallsystem energy savings without violating the performance constraints.A power management algorithm for the peripherals of Tianhe-1Ahigh-performance computer was proposed. According to the relationship between theload and temperature of CPU, an intelligent control strategy based on the perception offans was proposed strive to achieve precise cooling while reduce22.9%of the fanpower consumption. By using a distribution according to need strategy, power supplyunit can work efficient while achieve a power savings of17.5%.By using a power-aware resource management and Regulation of systemperipherals, the power management system proposed in this paper can achieveconsiderable energy saving for high-performance computing, which can be economiccost savings, but also has a good social effect.