边缘海静力数值预报模式并行算法研究  

Parallelization of Hydrostatic Numerical Forecasting Model of Marginal Sea

作　　者：逄仁波[1,2] 张云泉[3] 谭光明[3] 徐建良[1] 贾海鹏[3] 解庆春[4] 

机构地区：[1]中国海洋大学信息科学与工程学院,青岛266100 [2]国家海洋环境预报中心网络与计算机部,北京100081 [3]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,北京100190 [4]中国传媒大学高性能计算中心,北京100024

出　　处：《计算机科学》2016年第1期14-17,29,共5页Computer Science

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

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。The hydrostatic numerical forecasting model of marginal sea is the numerical forecasting model developed independently in China according to the characteristics of the marginal sea.There are many physical equations in the model,and some equations are not appropriate for parallelism like SOR,so it takes a long time to run the program.To solve these problems,the parallel SOR method,which is based on 3Dcomputing grids and characters of ocean forecasting models,is used to solve the parallel problem and keep dependencies between data in 3Dcomputing grids.The methods are also used to improve the efficiency of the parallel program,such as using MPI non-blocking communication,cutting the process of computing and communication to more steps,and overlapping the steps of communication with computing.The experiments show that the performance of the parallel hydrostatic numerical forecasting model increases 60.71 times compared to the serial program,and root mean square error of three-day forecasting results（25920time steps）is less than 0.001,which meets the numerical ocean forecasting requirements of timeliness and accuracy.

关 键 词：数值模式 预报 并行 非阻塞 SOR 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]