中尺度天气预报模式WRF在“元”上的性能评估

Performance evaluation of weather research and forecast(WRF)model on ERA

介绍新一代中尺度预报模式(WRF)的基本概况以及编译运行WRF时需要重点关注的几个问题;描述配置WRF性能测试的数值实验的方法,通过这组数值实验,借助Intel VTune以及PAPI性能测试工具,重点研究如何在高性能计算机上对WRF进行合理的处理器划分配置以及OpenMP线程配置,比较不同配置的并行效率,得到最优的配置方案;测试WRF的强扩展性、弱扩展性以及浮点计算效率。基于以上工作,对WRF在集群系统上的高效使用予以总结与建议。

The basic information of weather research and forecast model（WRF）was introduced,and some issues that should be paid extra attention to during compiling and running were mentioned.A series of numerical experiments with different processor divisions and OpenMP configurations were conducted.The performance was illustrated by performance application programming interface（PAPI）and Intel VTune.By comparing the parallel efficiency,the optimal strategy of processor division and OpenMP configuration on high performance computers was concluded.In addition,the strong and weak scalability of WRF were tested,as well as MFLOPS.Suggestions on how to apply WRF on cluster systems with high efficiency were provided based on the experiments conducted.