摘要
研究了CMAQ大气模型在64位Linux操作系统上不同CPU核心数目并行计算模拟耗时以及结果的差异情况。研究结果表明,并行计算能大幅缩短CMAQ模拟耗时,以16个CPU核心并行处理为性价比最佳值;此时连续模拟中国区域37天空气质量状况(分辨率36 km、167行×97列、垂直14层)平均耗时小于16 m in/d,而相同情况下单核模拟耗时大于2 h/d;多于16个核心并行处理时,随核心数量的增加模型性能提升的趋势减缓;操作系统和参与运算的核心数目对CMAQ模型模拟结果没有影响。
The computational efficiency and simulation results of community multiscale air quality （CMAQ） model developed by U. S. EPA were analyzed using a variety of parallel CPU cores under two 64 bit Linux OS platforms. The results show that the parallel computing can significantly reduce the CMAQ run time, and 16 appear to be the optimal performance comparing hardware investment parallel CPU cores number. The average simulation time per day was reduced to less than 16 minutes using the 16 parallel cores, compared to up to 2 hours using single core when the air quality were simulated by CMAQ for a 37- day period over a main Chinese national domain with a 36km resolution, 167 row × 97 column, 14 layers. The improvement of computational efficiency per CPU seemed to drop rapidly when more than 16 CPU cores were used. In addition, this paper verified that the simulation results were not be affected by the usage of different number of parallel CPU cores or different operation system.
出处
《计算机应用研究》
CSCD
北大核心
2009年第6期2266-2269,共4页
Application Research of Computers
基金
中美双边合作项目
广东省普通高等学校重点实验室开放基金资助项目
国家“863”计划资助项目(2006AA06A308)
关键词
并行计算
空气质量模型
CMAQ
LINUX
parallel computation
air quality modal
community multiscale air quality（CAMQ）
Linux
