计算机科学 ›› 2016, Vol. 43 ›› Issue (1): 14-17.doi: 10.11896/j.issn.1002-137X.2016.01.003

• 目次 • 上一篇    下一篇

边缘海静力数值预报模式并行算法研究

逄仁波,张云泉,谭光明,徐建良,贾海鹏,解庆春   

  1. 中国海洋大学信息科学与工程学院 青岛266100;国家海洋环境预报中心网络与计算机部 北京100081,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,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,中国海洋大学信息科学与工程学院 青岛266100,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,中国传媒大学高性能计算中心 北京100024
  • 出版日期:2018-12-01 发布日期:2018-12-01
  • 基金资助:
    本文受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

Parallelization of Hydrostatic Numerical Forecasting Model of Marginal Sea

PANG Ren-bo, ZHANG Yun-quan, TAN Guang-ming, XU Jian-liang, JIA Hai-peng and XIE Qing-chun   

  • Online:2018-12-01 Published:2018-12-01

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。

关键词: 数值模式,预报,并行,非阻塞,SOR

Abstract: 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 mo-del,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 3D computing grids and characters of ocean forecasting models,is used to solve the parallel problem and keep dependencies between data in 3D computing 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 (25920 time steps) is less than 0.001,which meets the numerical ocean forecasting requirements of timeliness and accuracy.

Key words: Numerical model,Forecasting,Parallelization,Non-blocking,SOR

[1] Hueske F,Markl V.Optimization of Massively Parallel DataFlows [M]∥Large-Scale Data Analytics,2014:41-74
[2] Wu Xing,Liu Yan.Optimization of Load Adaptive Distributed Stream Processing Services [C]∥Anchorage:Services Computing.2014:504-511
[3] Ren Xiao-guang,Zhou Wen-hao,Chen Juan.Collective Communication Optimization for Solving Linear Algebraic Equations [J].Advanced Materials Research,2014,989-994:4934-4939
[4] Liu Li,Li Rui-zhe,Yang Guang-wen,et al.Improving Parallel Performance of a Finite-Difference AGCM on Modern High-Performance Computers [J].Oceanic Technol.,2014(31):2157-2168
[5] Yan Chao-kun,Hu Zhi-gang,Li Xi,et al.Reliability-Cost Optimization Scheduling Model and Algorithm in Grid [J].Compu-ter Science,2013,40(3):136-141(in Chinese)阎朝坤,胡志刚,李玺,等.面向可靠性-费用优化的网格任务调度模型及算法研究 [J].计算机科学,2013,40(3):136-141
[6] Li Xin,Jia Zhi-ping,Ju Lei,et al.Energy Efficient Scheduling and Optimization for Parallel Tasks on Homogeneous Clusters [J].Chinese Journal of Computers,2012,35(3):591-602(in Chinese)李新,贾智平,鞠雷,等.一种面向同构集群系统的并行任务节能调度优化方法 [J].计算机学报,2012,35(3):591-602
[7] Cotronis Y,Konstantinidis E,Louka M A.A comparison of CPU and GPU implementations for solving the Convection Diffusion equation using the local Modified SOR method [J].Parallel Computing,2014,40(7):173-185
[8] Konstantinidis E,Cotronis Y.Graphics processing unit acceleration of the red/black SOR method[J].Concurrency and Computation,2013,25(8):1107-1120
[9] Epicoco I,Mocavero S,Aloisio G.The performance model for a parallel SOR algorithm using the red-black scheme [J].International Journal of High Performance Systems Architecture,2012,4(2):101-109
[10] Satoa Y,Hino T,Ohashi K.Parallelization of an unstructured Navier-Stokes solver using a multi-color ordering method for OpenMP [J].Computers & Fluids,2013,8(12):496-509
[11] Iwashita T,Nakashima H,Takahashi Y.Algebraic Block Multi-Color Ordering Method for Parallel Multi-Threaded SparseTriangular Solver in ICCG Method [C]∥Proceeding of IEEE 26th International Parallel and Distributed Processing Symposium.Shanghai,China,2012:474-483
[12] Wang Chuan-long,Meng Guo-yan,Yong Xue-rong.Modifiedparallel multisplitting iterative methods for non-Hermitian positive definite systems [J].Advances in Computational Mathematics,2013,38(4):859-872
[13] Meng Guo-yan,Wang Chuan-long,Yan Xi-hong.Self-adaptiveNon-stationary Parallel Multisplitting Two-Stage Iterative Methods for Linear Systems [J].Data and Knowledge Enginee-ring,2012,7696:38-47
[14] Zhang Cheng-yi,Luo Shuang-hua,Xu Zong-ben.On parallelmultisplitting block iterative methods for linear systems arising in the numerical solution of Euler equations [J].Journal of Computational and Applied Mathematics,2015,279:249-260
[15] Zhang Li-tao,Li Jian-lei,Gu Tong-xiang,et al.Convergence of Relaxed Matrix Parallel Multisplitting Chaotic Methods for Matrices [J].Journal of Applied Mathematics,2014,2014:1-9
[16] Zhang Li-li.Two-Stage Multisplitting Iteration Methods Using Modulus-Based Matrix Splitting as Inner Iteration for Linear Complementarity Problems [J].Journal of Optimization Theory and Applications,2014,160(1):189-203
[17] Zheng Ning,Yin Jun-feng.Accelerated modulus-based matrixsplitting iteration methods for linear complementarity problem [J].Numerical Algorithms,2013,64(2):245-262
[18] Bai Zhong-zhi,Zhang Li-li.Modulus-based synchronous two-stage multisplitting iteration methods for linear complementarity problems [J].Numerical Algorithms,2013,62(1):59-77
[19] Butrylo B,Tudruj M,Masko L.Parallel SSOR preconditioning implemented on dynamic SMP clusters with communication on the fly [J].Future Generation Computer Systems,2010,26(3):491-497
[20] Xu Qiu-yan.A New Parallel Successive Overrelaxation Iterative Algorithm for Poisson Equation [C]∥Proceeding of 2011 International Conference on Cyber-Enabled Distributed Computing and Knowledge Discovery.Beijing,China,2011:295-300

No related articles found!
Viewed
Full text


Abstract

Cited

  Shared   
  Discussed   
No Suggested Reading articles found!