通过数值试验分析了区域分解策略对ARPS气象模式并行计算速度的影响,发现无论是否使用编译优化技术,均以分解后数据区域近似为正方形时具有最大的加速比和并行效率。在二级编译优化的情况下,并行速度还和分解方向有关,在Y方向上的分解比在X方向上的分解更有利于提高并行效率,而在无优化情况下,并行速度和分解方向几乎无关。并从通信量和编译优化的角度对试验结果进行了讨论和分析。
The influence of domain decomposition on the parallel computation speed of ARPS model is analyzed on the base of some experiments. Results show that the subdomalns similar to square possess the biggest acceleration rate and efficiency of parallel whether the optimization lever -o2 is applied to or not. Under the circumstances of optimization lever -o2, parallel speed also relates to decomposition direction. The decomposition along y direction is preferable to that along x direction for improving in parallel efficiency. However, under the circumstances without optimization, parallel speed is almost independent of decomposition direction. Furthermore, experimental results are discussed in the paper from the angles of communications traffic and optimization.