基于神威加速计算架构的LBM多级并行计算
作者:
基金项目:

高等学校学科创新引智计划(B23008); 未来网络科研基金(FNSRFP2021YB11)


LBM Multi-level Parallel Computing Based on SACA
Author:
  • 摘要
  • | |
  • 访问统计
  • |
  • 参考文献 [15]
  • |
  • 相似文献 [20]
  • | | |
  • 文章评论
    摘要:

    格子玻尔兹曼方法(lattice Boltzmann method, LBM)是一种基于分子运动理论计算流体力学(computational fluid dynamics, CFD)的方法, 提高LBM的并行计算能力是高性能计算领域的一项重要的研究内容. 本文基于SW26010Pro处理器, 通过区域分解、数据重构、双缓冲、向量化等优化方法, 实现了LBM的多级并行. 基于以上优化方案, 测试了5 600万网格规模, 实现结果显示, 相比于MPI进行级并行, 碰撞过程的平均加速倍数达到61.737、迁移过程的平均加速倍数达到17.3, 同时对方腔流案例做了强扩展测试, 网格规模为1200×1200×1200, 以6.2万计算核心为基准, 百万核心的并行效率超过60.5%.

    Abstract:

    The lattice Boltzmann method (LBM) is a computational fluid dynamics (CFD) method based on molecular motion theory. Improving the parallel computing capability of LBM is an important research topic in the high-performance computing field. This article is based on the SW26010Pro processor and achieves multi-level parallelism of LBM through optimization methods such as region decomposition, data reconstruction, double buffering, and vectorization. Based on the above optimization methods, a grid size of 56 million is tested, and the implementation results show that compared to message passing interface (MPI) level parallelism, the average acceleration factor of the collision process reaches 61.737, and that of the migration process reaches 17.3. At the same time, strong expansion testing is conducted on the lid-driven cavity flow case, with a grid size of 1200×1200×1200. Based on 62 000 computing cores, the parallel efficiency of one million cores exceeds 60.5%.

    参考文献
    [1] Vinuesa R, Brunton SL. Enhancing computational fluid dynamics with machine learning. Nature Computational Science, 2022, 2(6): 358–366.
    [2] Astoul T, Wissocq G, Boussuge JF, et al. Lattice Boltzmann method for computational aeroacoustics on non-uniform meshes: A direct grid coupling approach. Journal of Computational Physics, 2021, 447: 110667.
    [3] Chaaban M, Heider Y, Markert B. A multiscale LBM-TPM-PFM approach for modeling of multiphase fluid flow in fractured porous media. International Journal for Numerical and Analytical Methods in Geomechanics, 2022, 46(14): 2698–2724.
    [4] Phillips JC, Hardy DJ, Maia JDC, et al. Scalable molecular dynamics on CPU and GPU architectures with NAMD. The Journal of Chemical Physics, 2020, 153(4): 044130.
    [5] De Luca P, Galletti A, Ghehsareh HR, et al. A GPU-CUDA framework for solving a two-dimensional inverse anomalous diffusion problem. Parallel Computing: Technology Trends. IOS Press, 2020. 311.
    [6] Takizawa H, Shiotsuki S, Ebata N, et al. OpenCL-like offloading with metaprogramming for SX-Aurora TSUBASA. Parallel Computing, 2021, 102: 102754.
    [7] Li D, Xu CF, Wang YX, et al. Parallelizing and optimizing large-scale 3D multi-phase flow simulations on the Tianhe-2 supercomputer. Concurrency and Computation: Practice and Experience, 2016, 28(5): 1678–1692.
    [8] 丁越, 徐传福, 邱昊中, 等. 基于SYCL的多相流LBM模拟跨平台异构并行计算研究. 计算机科学, 2023, 50(11): 32–40.
    [9] 柳安军, 殷洪辉, 王利, 等. 基于新一代神威超算的计算流体力学Palabos软件的并行优化. 计算机科学, 2022, 49(10): 66–73.
    [10] Duan XH, Zhang M, Liu WG, et al. Tuning a general purpose software cache library for TaihuLight’s SW26010 processor. CCF Transactions on High Performance Computing, 2020, 2(2): 164–182.
    [11] 何强, 李永健, 黄伟峰, 等. 基于MPI+OpenMP混合编程模式的大规模颗粒两相流LBM并行模拟. 清华大学学报(自然科学版), 2019, 59(10): 847–853.
    [12] 陶颖军. 基于OpenCV的人脸识别应用. 计算机系统应用, 2012, 21(3): 220–223.
    [13] 王振华, 张鑫月, 刘智翔, 等. 遥感地物分割的改进格子玻尔兹曼并行模型. 遥感信息, 2021, 36(4): 1–6.
    [14] 吕小敬, 刘钊, 褚学森, 等. 面向超大规模并行模拟的LBM计算流体力学软件. 计算机科学, 2020, 47(4): 13–17.
    [15] 莫尚丰, 周振芬, 胡勇华, 等. 基于FT-M7002的复数域行向量矩阵乘法移植与优化. 计算机科学, 2023, 50(S2): 839–844.
    引证文献
    网友评论
    网友评论
    分享到微博
    发 布
引用本文

王鑫,张祖雨.基于神威加速计算架构的LBM多级并行计算.计算机系统应用,2024,33(8):60-67

复制
分享
文章指标
  • 点击次数:503
  • 下载次数: 1187
  • HTML阅读次数: 795
  • 引用次数: 0
历史
  • 收稿日期:2024-01-31
  • 最后修改日期:2024-02-29
  • 在线发布日期: 2024-06-28
文章二维码
您是第12821914位访问者
版权所有:中国科学院软件研究所 京ICP备05046678号-3
地址:北京海淀区中关村南四街4号 中科院软件园区 7号楼305房间,邮政编码:100190
电话:010-62661041 传真: Email:csa (a) iscas.ac.cn
技术支持:北京勤云科技发展有限公司

京公网安备 11040202500063号