WANG Ya-ru,WANG Peng,WANG De-zhi.Performance Test and Analysis of Multi-core Parallel Mode based on MPI[J].Journal of Chengdu University of Information Technology,2018,(06):617-623.[doi:10.16836/j.cnki.jcuit.2018.06.004]
基于MPI的多核并行模式的性能测试与分析
- Title:
- Performance Test and Analysis of Multi-core Parallel Mode based on MPI
- 文章编号:
- 2096-1618(2018)06-0617-07
- Keywords:
- parallel algorithm; Monte Carlo; speed up; efficiency; computing for communicating; parallel computing program flow chart
- 分类号:
- TP302.7
- 文献标志码:
- A
- 摘要:
- 为了对蒙特卡洛并行程序进行有效的表达,提出一种并行计算程序流程图的表示方法,并行算法在实际应用中对效率制约的瓶颈就是进程间的通信过程。在“多计算、少通信”与“少计算、多通信”两种不同并行模式下,通过控制集群节点数以及节点核数来测试基于MPI实现的蒙特卡洛算法的加速比与效率,分析并行程序在多核并行模式下的计算性能。结果表明采取“计算换通信”的策略能有效地实现对算法的并行加速,提高了系统的可扩展性,为今后提升并行编程效率提供了参考,弥补了传统的串行程序流程图无法表述多进程之间信息沟通及协作的不足。
- Abstract:
- In order to express the Monte Carlo parallel program effectively, an expressing method of the parallel computing program flowchart is proposed, It's the communication between processes that restricts the efficiency of parallel algorithms in practical applications. There are two communicating parallel modes, which is “more-computing, less-communicating” and “less-computing and more-communicating”. In order to analyze the computation performance of parallel programs in multi-core and the two communicating parallel modes, the Monte Carlo algorithm based on MPI is used to test the speedup and efficiency by controlling the number of cluster nodes and the cores within node. The results show that the strategy of “computing for communicating” can effectively speedup the paralleling acceleration of the algorithm, improve the scalability of the system and provide reference for improving the efficiency of parallel programming in the future.which makes up for the lack of the traditional serial program flow chart that can't express the information communication and cooperation between multiple processes.
参考文献/References:
[1] 王鹏,黄焱,安俊秀.云计算与大数据技术[M].北京:人民邮电出版社,2014:66-67.
[2] 王鹏.云计算的关键技术与应用实例[M].北京:人民邮电出版社,2010:29-31.
[3] Shubhangi Rastogi,Hira Zaheer.Significance of Parallel Computation over Serial Computation Using OpenMP,MPI,and CUDA[J].Quality,IT and Business Operations,2017:359-367.
[4] 罗红兵,张晓霞.MPI集合通信性能可扩展性研究与分析[J].计算机科学与探索,2017,11(2):252-261.
[5] 米东,施小清.基于MPI的TOUGHREACT并行性能分析[J].水文地质工程地质,2016,43(1):34-40.
[6] 谢超,麦联叨,都志辉.关于并行计算系统中加速比的研究与分析[J].计算机工程与应用,2002,26(3):66-68.
[7] 刘文志.并行算法设计与性能优化[M].北京:机械工业出版社,2015:110-130.
[8] 高珂,陈荔城,范东睿.多核系统共享内存资源非配和管理研究[J].计算机学报,2015,38(5):1020-1034.
[9] 周伟.并行计算在海洋水龄谱模拟中的应用研究[D].邯郸:河北工程大学,2013:18-19.
[10] 康辉,王家琦.基于PI演算的并行编程语言[J].吉林大学学报(工学版),2016,46(1):235-241.
[11] Yang C T,Huang C L,Lin C F.Hybrid CUDA,OpenMP,and MPI parallel programming on multicore GPU clusters[J].Computer Physics Communications,2011,182(1):266-269.
[12] 王蕾,崔慧敏.任务并行编程模型研究与进展[J].软件学报,2013,24(1):77-90.
备注/Memo
收稿日期:2018-06-05 基金项目:国家自然科学基金资助项目(60702075); 四川省青年科学基金资助项目(09ZQ026-068)