边缘海静力数值预报模式并行算法研究

发布时间：2018-05-01 18:13

本文选题：数值模式 + 预报　；参考：《计算机科学》2016年01期

【摘要】：边缘海静力数值模式是国内针对边缘海特点自主开发的数值预报模式,但该模式因物理求解方程较多且采用不宜并行化的SOR求解算法而程序计算时间过长。针对上述问题,提出基于三维网格和海洋模式特点的SOR并行求解算法,该算法在保留三维网格数据间依赖关系的同时,有效解决了SOR迭代算法难以并行化的问题。同时,引入通信避免算法,采用MPI非阻塞通信方式,细分计算和通信过程,利用计算有效隐藏通信开销,提高了并行程序效率。实验结果表明,并行后的边缘海静力数值模式程序的性能相对串行程序提升了60.71倍,3天(25920计算时间步)预报结果的均方根误差低于0.001,满足海洋数值预报的时效性和精度要求。
[Abstract]:The boundary sea static numerical model is a numerical prediction model developed independently in China according to the characteristics of the edge sea. However, due to the large number of physical solving equations and the unparallelism of the SOR algorithm, the program computation time of the model is too long. In order to solve the above problems, a parallel SOR algorithm based on the characteristics of 3D mesh and ocean model is proposed. This algorithm can effectively solve the problem that SOR iteration algorithm is difficult to parallelize while preserving the dependency relationship between 3D grid data. At the same time, the communication avoidance algorithm is introduced, the MPI non-blocking communication mode is adopted, the computation is used to subdivide the computation and the communication process, and the computation is used to hide the communication overhead effectively, and the efficiency of the parallel program is improved. The experimental results show that the performance of the parallel program is 60.71 times higher than that of the serial program. The root-mean-square error of the prediction results is less than 0.001, which meets the requirements of the time efficiency and precision of the ocean numerical prediction.
【作者单位】：中国海洋大学信息科学与工程学院;国家海洋环境预报中心网络与计算机部;中国科学院计算技术研究所计算机体系结构国家重点实验室;中国传媒大学高性能计算中心;
【基金】：海洋公益性行业科研专项:海洋渔业安全环境保障服务系统关键技术研究及示范应用(201205006) 国家自然科学基金面上项目(61272136) 国家自然基金重点基金(61133005),国家自然基金创新群体成员(60921002) 中国科学院计算技术研究所创新课题(20136030),中国科学院计算技术研究所计算机体系结构国家重点实验室创新课题(CARCH301)资助
【分类号】：P731.3;TP301.6

【相似文献】