Direct numerical simulation of flow around a 3D finite square cylinder using the Sunway Taihu Light
-
摘要:
目的 旨在探索基于国产处理器的异构超算平台在船舶水动力学领域的应用效果。 方法 基于“神威·太湖之光”超级计算机,采用MPI+Athread的编程方法,对雷诺数Re=250的三维有限长方柱绕流进行直接数值模拟,并对模拟结果进行验证与分析。模拟使用的网格规模最大达到245.76百万(t=600 s,dt=0.001),并行规模最高达到133 120核。 结果 经统计,在133 120核并行规模下245.76百万网格规模计算能够在数天之内完成。模拟结果显示,在三维有限长方柱绕流流动中,方柱各横截面具有同步涡脱的特征;对比不同长径比方柱绕流尾流场,发现长径比为2时的尾流涡系结构呈现出长直状的流向涡二次结构,而大于2时则为反对称卡门涡。 结论 模拟表明,基于“神威·太湖之光”超级计算机的多级并行计算可有效减少小尺度网格下因规模提升所导致的时间成本,在船舶水动力学领域有较好的应用潜力。 -
关键词:
- 三维有限长方柱绕流 /
- 直接数值模拟 /
- 并行计算 /
- “神威·太湖之光”超级计算机
Abstract:Objective This paper aims to study the application effects of heterogeneous supercomputing platforms based on domestic processors in the field of ship hydrodynamics. Methods The direct numerical simulation of the flow around a 3D finite square cylinder with Re = 250 using the Sunway Taihu Light supercomputer is implemented, in which the maximum number of grids is 245.76 million ( t = 600 s, dt = 0.001), and the simulation results are analyzed and verified. The parallel programming method of MPI+Threads is used with a maximum parallel size of 133 120 cores. Results According to the statistics, the calculation can be completed in a few days under the current parallel scale using 245.76 million grids. In addition, the simulation results show that the vortex shedding is synchronous in different cross-sections of a 3D finite square cylinder. When the characteristics of flow around the finite square cylinder with slenderness ratios of 2, 3 and 4 are compared, it is found that when the slenderness ratio is 2, the wake vortex structure of the wake is a long straight streamwise vortex secondary structure, or else it is an antisymmetric Karman vortex. Conclusion These results indicate that multilevel parallel computing using the Sunway Taihu Light supercomputer can effectively reduce the time-consumption caused by grid scale increase in small-scale grids, which has broad application prospects in the field of ship hydrodynamics. -
表 1 3D顶板驱动方腔流MPI / MPI+Athread并行效率
Table 1. The parallel efficiency of MPI/MPI+Athread for 3D cavity driven flow
网格数/百万 MPI / MPI+Athread/% 进程数4 进程数16 进程数64 进程数128 1
(100×100×100)93.2/84.88 87.14/53.35 73.07/28.50 50.35/14.78 3.375
(150×150×150)95.37/87.57 89.68/55.23 77.39/34.30 65.42/20.25 8
(200×200×200)96.35/90.84 91.24/59.46 84.31/40.48 73.30/24.16 15.625
(250×250×250)99.94/91.34 95.47/62.95 87.03/40.82 85.98/34.24 注:并行效率计算参照对象:MPI(单个主核计算耗时);MPI+Athread(单个核组耗时:主核+从核组) 表 2 方柱绕流数值模拟的相关参数
Table 2. The related parameters of simulation for the flow around square cylinder
网格数/百万 网格
分辨率时间
步长/sRe 核数 迭代
步数/s耗时/h 3.84 (240×160×100) 0.1d 0.001 250 33 280 6×105 17 30.72 (480×320×200) 0.05d 33 280 79 245.76 (960×640×300) 0.025d 133 120 160 表 3 不同网格规模下方柱绕流数值模拟的流场特征频率
Table 3. The characteristic frequency of different grid sizes for flow around a square cylinder
网格数/百万 x=15d x=22d 主频/Hz 主频幅值/dB 主频/Hz 主频幅值/dB 3.84 0.129 −15.96 0.129 −16.46 30.72 0.130 −14.08 0.130 −15.70 245.76 0.130 −13.86 0.130 −14.93 表 4 不同长径比方柱绕流流动平均阻力系数及斯特劳哈尔数
Table 4. The average flow resistance coefficient and Strouhal number of the flow around a 3D finite square cylinder with different slenderness ratio
算例 H/d St Cd 本文计算 2 0.095 1.075 3 0.115 1.18 4 0.130 1.245 文献[10] 2 0.097 1.08 3 0.114 1.16 4 0.124 1.23 -
[1] 张云泉, 袁良, 袁国兴, 等. 2020年中国高性能计算机发展现状分析与展望[J]. 数据与计算发展前沿, 2020, 2(6): 1–10.ZHANG Y Q, YUAN L, YUAN G X, et al. State-of-the-art analysis and perspectives of China HPC development in 2020[J]. Frontiers of Data & Computing, 2020, 2(6): 1–10 (in Chinese). [2] 李燕. 基于神威太湖之光的稠密矩阵特征值求解算法及性能优化[D]. 北京: 中国科学院大学, 2018.LI Y. Dense matrix eigenvalue solving algorithm and performance optimization based on Sunway Taihulight[D]. Beijing: University of Chinese Academy of Sciences, 2018 (in Chinese). [3] 李芳, 李志辉, 徐金秀, 等. 基于十亿亿次国产超算系统的流体力学软件众核适应性研究[J]. 计算机科学, 2020, 47(1): 24–30. doi: 10.11896/jsjkx.181102176LI F, LI Z H, XU J X, et al. Research on adaptation of CFD software based on many-core architecture of 100P domestic supercomputing system[J]. Computer Science, 2020, 47(1): 24–30 (in Chinese). doi: 10.11896/jsjkx.181102176 [4] 张亚英, 吴乘胜, 王建春, 等. 面向众核处理器的水动力学CFD并行计算探索[J]. 船舶, 2021, 32(4): 15–23.ZHANG Y Y, WU C S, WANG J C, et al. Multi-core processor oriented parallel computation for CFD in hydrodynamics[J]. Ship & Boat, 2021, 32(4): 15–23 (in Chinese). [5] SAKAMOTO H, ARIE M. Vortex shedding from a rectangular prism and a circular cylinder placed vertically in a turbulent boundary layer[J]. Journal of Fluid Mechanics, 1983, 126: 147–165. doi: 10.1017/S0022112083000087 [6] WANG H F, ZHOU Y. The finite-length square cylinder near wake[J]. Journal of Fluid Mechanics, 2009, 638: 453–490. doi: 10.1017/S0022112009990693 [7] 陶文铨. 数值传热学[M]. 2版. 西安: 西安交通大学出版社, 2001.TAO W Q. Numerical Heat Transfer[M]. 2nd ed. Xi'an: Xi'an Jiaotong University Press, 2001 (in Chinese). [8] HUNT J C R, WRAY A A, MOIN P. Eddies, streams, and convergence zones in turbulent flows: N89-24555[R]. Stanford: Center for Turbulence Research, 1988: 193-208. [9] 高健停. 三维柱体绕流自由端效应的数值研究[D]. 哈尔滨: 哈尔滨工程大学, 2019.GAO J T. Numerical study on free end effect of flow around a cylinder[D]. Harbin: Harbin Engineering University, 2019 (in Chinese). [10] SAHA A K. Unsteady flow past a finite square cylinder mounted on a wall at low Reynolds number[J]. Computers & Fluids, 2013, 88: 599–615. -
ZG2595_en.pdf
-