摘要：Energy consumption has become a serious problem in high-performance computing (HPC) systems. Parallel loops often occupy a significant part of the execution time of overall parallel programs. Thus, reducing their energy consumption is the key to the reduction in energy consumption of the program. This paper discusses energy optimization in OpenMP loop scheduling, which is a method of optimizing energy consumption of parallel loops within a certain performance loss. OpenMP programming includes Uniform-Sized scheduling unit (US)-based Energy-saving Optimal Static Scheduling algorithm (EOSS), UnUniform-Sized scheduling unit (UUS)-based Energy-oriented Static Scheduling algorithm (ESS). EOSS obtains the maximum energy savings under the ideal condition (US) by combining loop rescheduling and voltage/frequency scaling. ESS optimizes energy consumption under the non-ideal condition (UUS). We evaluate these energy-oriented loop scheduling algorithms through simulations on a multiprocessor system. Taking 160 processors as example, five NPB programs with EOSS can reduce 43.5% energy consumption on the average, which is 2.1% more energy saved than the DVFS-only method. ESS can achieve 54% energy savings with the example code we use.
关键词：OpenMP loop scheduling;dynamic voltage frequency scaling;processor shutdown;energy optimization
