国产众核处理器百万核时域有限差分并行计算

doi:10.3969/j.issn.1001-2400.2017.06.012

西安电子科技大学学报

国产众核处理器百万核时域有限差分并行计算

姜雪松;江树刚;张玉;赵勋旺

(西安电子科技大学天线与微波技术重点实验室，陕西西安 710071)

收稿日期:2016-12-11 出版日期:2017-12-20 发布日期:2018-01-18
作者简介:姜雪松(1991-)，男，西安电子科技大学硕士研究生，E-mail：1530533866@qq.com
基金资助:
国家重点研究开发计划资助项目(2017YFB0202102)；国家重点研发计划国际科技合作资助项目(2016YFE0121600)；教育部新世纪优秀人才支持计划资助项目(NCET-13-0949)；国家高技术研究发展计划(863 计划)资助项目(2012AA01A308，2014AA01A302)

Parallel FDTD computation of million cores with the domestically-made many-core supercomputer

JIANG Xuesong;JIANG Shugang;ZHANG Yu;ZHAO Xunwang

(Science and Technology on Antennas and Microwave Lab., Xidian Univ., Xi'an 710071, China)

Received:2016-12-11 Online:2017-12-20 Published:2018-01-18

摘要/Abstract

摘要：

根据“神威太湖之光”超级计算机新型架构，研究了众核计算平台超大规模并行时域有限差分方法．利用太湖之光提供的加速线程库，提出一种针对众核处理器的时域有限差分并行编程方法．基于该方法设计的并行时域有限差分程序，率先突破300万处理器核并行计算，以16384核为基准，在327680核时并行效率可达50％．并完成200波长电尺寸机载天线受扰分析．研究结果表明，并行时域有限差分法可以在国产众核超级计算平台上解决复杂电磁工程问题．

关键词: 并行算法, 国产处理器核, 300万处理器核, 时域有限差分法, 计算电磁学

Abstract:

According to the new structure of domestically-made super computers, the massively parallel Finite-Difference Time-Domain (FDTD) is studied. By using the accelerating thread library, a parallel FDTD programming method for many-core processors is proposed. The parallel FDTD program based on this method is the first to realize 3 million processor cores parallel computation.Test results show that the parallel efficiency can reach up to 50％ on 327680 cores with 16384 cores as the benchmark. And an analysis of the 200 wavelength electric size airborne antenna array is made. The researches in this paper indicate that the complicate electromagnetic problems can be solved efficiently using the method on the domestic many-core supercomputing platform.

Key words: parallel algorithms, domestic processors, 3000000 core, finite-difference-time-domain, computational electromagnetics

姜雪松;江树刚;张玉;赵勋旺. 国产众核处理器百万核时域有限差分并行计算[J]. 西安电子科技大学学报, 2017, 44(6): 65-69+128.

JIANG Xuesong;JIANG Shugang;ZHANG Yu;ZHAO Xunwang. Parallel FDTD computation of million cores with the domestically-made many-core supercomputer[J]. Journal of Xidian University, 2017, 44(6): 65-69+128.

参考文献

［1］ YEE K S. Numerical Solution of Initial Boundary Value Problems Involving Maxwell's Equations in Isotropic Media［J］. IEEE Transactions on Antennas and Propagation, 1966, 14(3): 302-307.
［2］葛德彪, 闫玉波. 电磁波时域有限差分方法［M］. 3版. 西安: 西安电子科技大学出版社, 2011.
［3］张玉. 电磁场并行计算［M］. 西安: 西安电子科技大学出版社, 2006.
［4］张立红, 余文华, 杨小玲. 加速并行时域有限差分仿真的新方法［J］. 电波科学学报, 2012, 27(1): 56-60.
ZHANG Lihong, YU Wenhua, YANG Xiaoling. New Acceleration Technique for Parallel FDTD Simulation［J］. Chinese Journal of Radio Science, 2012, 27(1): 56-60.
［5］ BATISTA C, DO REGO C, NUNES M, et al. Improved High-order FDTD Parallel Propagator for Realistic Urban Scenarios and Atmospheric Conditions［J］. IEEE Antennas and Wireless Propagation Letters, 2016, 15: 1779-1782.
［6］ ZHANG X, DUAN H, JIANG T. Research on Rough Sea Surface EM Computation Model Based on Parallel FDTD Method［C］//Progress in Electromagnetic Research Symposium. Piscataway: IEEE, 2016: 2193-2193.
［7］ WAIDYASOORIYA H M, HARIYAMA M, OHTERA Y. FPGA Architecture for 3-D FDTD Acceleration Using OpenCL［C］//Progress in Electromagnetic Research Symposium. Piscataway: IEEE, 2016: 4719.
［8］ PECQUEUX B, LABARBE L. Software and Hardware Assessment of FDTD Simulations for Very Large and Complex Scenes［C］//Proceedings of the IEEE International Symposium on Electromagnetic Compatibility. Piscataway: IEEE, 2016: 640-643.
［9］ PARK S M, KIM E K, PARK Y B, et al. Parallel Dispersive FDTD Method Based on the Quadratic Complex Rational Function［J］. IEEE Antennas and Wireless Propagation Letters, 2016, 15: 425-428.
［10］ WAIDYASOORIYA H M, HARIYAMA M. FPGA-based Deep-pipelined Architecture for FDTD Acceleration Using OpenCL［C］//Proceedings of the 2016 IEEE/ACIS 15th International Conference on Computer and Information Science. Piscataway: IEEE, 2016: 7550742.
［11］ WAIDYASOORIYA H M, TAKEI Y, TATSUMI S, et al. OpenCL-based FPGA-platform for Stencil Computation and Its Optimization Methodology［J］. IEEE Transactions on Parallel and Distributed Systems, 2017, 28(5): 1390-1402.
［12］江树刚, 林中朝, 张玉, 等. 国产超级计算机实现10万核FDTD并行计算［J］. 西安: 西安电子科技大学学报, 2015, 42(5): 86-91.
JIANG Shugang, LIN Zhongchao, ZHANG Yu, et al. Parallel FDTD Computation of Hundred Thousand Cores on a Domestically-made Supercomputer［J］. Journal of Xidian University, 2015, 42(5): 86-91.
［13］ JIANG S G, ZHANG Y, LIN Z C, et al. An Optimized Parallel FDTD Topology for Challenging Electromagnetic Simulations on Supercomputers［J］. International Journal of Antennas and Propagation, 2015, 2015(2015): 690510.
［14］唐川, 姜禾, 张勐, 等. 高性能计算发展动态与趋势观察［EB/OL］. ［2016-10-12］. http: //www. doc88.com/p-51480294334.html.

[1]	苏秦, 顾宗静, 林中朝, 赵勋旺, 张玉. 并行区域分解法分析千波长目标散射特性[J]. 西安电子科技大学学报, 2018, 45(6): 7-13.
[2]	张玉强;葛德彪. 一种高稳定的复杂色散介质分析FDTD方法[J]. 西安电子科技大学学报, 2018, 45(4): 118-122+173.
[3]	江树刚;林中朝;张玉;魏兵;赵勋旺. 国产超级计算机实现10万核FDTD并行计算[J]. J4, 2015, 42(5): 86-91.
[4]	王永;林中朝;张玉;梁昌洪. 并行矩量法分析波导缝隙天线的辐射特性[J]. J4, 2015, 42(5): 75-79+132.
[5]	林中朝;陈岩;张玉;江树刚;赵勋旺;吕兆峰. 国产CPU平台中并行高阶矩量法研究[J]. J4, 2015, 42(3): 43-47.
[6]	白冰;牛中奇. 时域有限差分法中的GPU加速高效CPML方案[J]. J4, 2015, 42(1): 194-199+212.
[7]	吴慧欣;王秉;柴争义. 一种采用并行免疫优化的频谱分配算法[J]. J4, 2014, 41(4): 111-115+136.
[8]	雷继兆;梁昌洪;张玉. 并行FDTD结合服务器分析电大电磁问题[J]. J4, 2009, 36(5): 846-850.
[9]	丁海1;褚庆昕2. TE模式下子域二阶精度FDTD算法 [J]. J4, 2009, 36(1): 162-165.
[10]	朱虎明;焦李成. 并行免疫克隆特征选择算法 [J]. J4, 2008, 35(5): 853-857.
[11]	陈智慧1;褚庆昕2. 任意线性集总网络FDTD建模新方法 [J]. J4, 2008, 35(2): 262-266.
[12]	杨凌霞;葛德彪;姜彦南. TDPO及其并行算法在电磁散射中的应用 [J]. J4, 2007, 34(7): 128-131.
[13]	王林年;褚庆昕;梁昌洪. 交替方向隐式时域有限差分法中的Berenger理想匹配层[J]. J4, 2006, 33(3): 458-461.
[14]	陈霖;刘其中;纪奕才. 广义正交坐标系中FDTD算法的APML吸收边界[J]. J4, 2004, 31(5): 770-773.
[15]	赵建勋1;鲁德强2. 细胞培皿构形对毫米波辐照剂量的影响[J]. J4, 2004, 31(1): 115-118.

国产众核处理器百万核时域有限差分并行计算

Parallel FDTD computation of million cores with the domestically-made many-core supercomputer

PDF (PC)

赞

可视化

被引次数

摘要/Abstract

引用本文

使用本文

参考文献

相关文章 15

Metrics

本文评价

推荐阅读 10