| 高精度湍流直接数值模拟程序的异构并行优化分析 |
| |
| 作者姓名： | 张天文 李新亮 张鉴 陆忠华 |
| |
| 作者单位： | 1. 中国科学院计算机网络信息中心, 北京 100190;
2. 中国科学院力学研究所, 北京 100190;
3. 中国科学院大学, 北京 100049 |
| |
| 摘 要： | 在众核处理器应用中,主要难点在于异构并行应用模式和负载均衡的策略,对于计算流体力学,需要针对相关应用设计相应的方案。我们针对湍流直接数值模拟中串行程序含有部分并行度较高的子程序或函数的特点,设计了一种新的并行计算模式,给出了一种异构平台优化方案,并在中科院超级计算系统"元"上进行了测试和分析,对领域内的典型算例进行了性能测试,着重讨论了不同规模下采用offload模式的CPU和MIC异构并行的扩展性能。
|
| 关 键 词： | 高性能计算 优化分析 异构并行 MIC |
| 收稿时间： | 2015-07-20 |
Optimization Analysis of the Heterogeneous Parallel Algorithm for High Accuracy Turbulence Direct Numerical Simulations |
| |
| Authors: | Zhang Tianwen Li Xinliang Zhang Jian Lu Zhonghua |
| |
| Affiliation: | 1. Supercomputing Center, Computer Network Information Center, Chinese Academy of Sciences, Beijing 100190, China;
2. Institute of Mechanics, Chinese Academy of Sciences, Beijing 100190, China;
3. University of Chinese Academy of Sciences, Beijing 100049, China |
| |
| Abstract: | In the application of many-core processors, the main difficulties are to choose the heterogeneous parallel application mode and the load balancing strategy.It is important to design appropriate solutions for relevant applications in high-performance computing of computational fluid dynamics.In allusion to the feature of that the serial program contains the parts of subprograms or functions with higher parallal performance in turbulence direct numerical simulations, we designed a new parallel computing model, and proposed a optimization scheme, and realized it on the supercomputing system, ERA, at CNIC, CAS.The performance test and analysis for typical computing examples were conducted, and the discussions are focused mainly on the scalability of the heterogeneous parallel algorithms with offload mode on the CPU and MIC heterogeneous architecture. |
| |
| Keywords: | high performance computing optimization analysis heterogeneous parallel MIC |
|
| 点击此处可从《》浏览原始摘要信息 |
|
点击此处可从《》下载全文 |
|
