邱剑,顾兆林.利用高阶分区并行算法实现直接数值模拟[J].计算力学学报,2008,25(1):20~24 |
| 码上扫一扫! |
利用高阶分区并行算法实现直接数值模拟 |
Implementation of direct numerical simulation by high order zone-divided parallel algorithm |
修订日期:2006-01-21 |
DOI:10.7511/jslx20081007 |
中文关键词: 直接数值模拟,混合层,并行计算,MPI |
英文关键词:DNS,mixing layer,parallel computation,MPI |
基金项目:国家重点基础研究发展计划(973计划)
,
国家重点基础研究发展计划(973计划)
,
教育部跨世纪优秀人才计划的联合资助项目 |
邱剑 顾兆林 |
西安交通大学能动学院,西安交通大学人居学院 西安710049 |
摘要点击次数: 1624 |
全文下载次数: 26 |
中文摘要: |
在基于MPI环境的分布式内存机群上,结合高阶WENO-RF格式的特点,实现了5阶WENO-RF格式的分区并行计算方法,计算精度不受分区和节点数量影响。使用该分区并行算法以三维可压缩时间发展混合层为例进行了直接数值模拟,验证了并行算法的准确性,表明机群并行运算可以显著扩展微机的计算能力,并行效率高,减少了计算的墙上时间,适合在小型高速局域网内进行大规模数值模拟计算。 |
英文摘要: |
A high order zone-divided parallel computation for direct numerical simulation(DNS) was implemented on a PC cluster system based on MPI platform,combined with fifth order WENO-RF scheme.Details of the way to deal with communication and virtual nodes were discussed to ensure that the number of nodes and zone-dividing method were irrelevant to accuracy of DNS.The parallel code was applied to DNS of three-dimensional compressible temporal mixing layer and analyses on simulation solutions,and indicated that the results were satisfactory and coincident with former researches well.The computation results are independent of the inner boundaries when virtual nodes for data buffer between nodes meet the requirements of the stencil for WENO-RF scheme.It reveals that distributed parallel computing with the zone-divided algorithm can achieve high parallel efficiency and speedup factor,helps to reduce hardware requirements for DNS,and has the ability to deal with large-scale direct numerical simulations on PC clusters. |
查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|
|