数据与计算发展前沿 ›› 2025, Vol. 7 ›› Issue (5): 3-15.

CSTR: 32002.14.jfdc.CN10-1649/TP.2025.05.001

doi: 10.11871/jfdc.issn.2096-742X.2025.05.001

• 专刊:国产算力新力量,助力科学计算应用新发展 • 上一篇    下一篇

三维连续-非连续单元法分布式计算策略

黄俊光1,2(),冯春2,*(),张一鸣3   

  1. 1.河北工业大学土木与交通学院,天津 300401
    2.中国科学院力学研究所,北京 100190
    3.浙江理工大学建筑工程学院,浙江 杭州 310018
  • 收稿日期:2025-03-19 出版日期:2025-10-20 发布日期:2025-10-23
  • 通讯作者: 冯春
  • 作者简介:黄俊光,河北工业大学,博士研究生,主要研究方向为高性能非线性数值计算、智能监控与自动化。
    本文主要承担工作为:模型建立、代码编程、算法实现和论文写作。
    HUANG Junguang is a Ph.D. candidate at Hebei University of Technology. wHis primary research interests include high-performance nonlinear numerical computation, intelligent monitoring, and automation.
    In this study, he is primarily responsible for model establishment, code programming, algorithm implementation, and paper writing.
    E-mail: 202111601004@stu.hebut.edu.cn|冯春,中国科学院力学研究所,高级工程师,历时10余年,主导开发了CDEM求解器、块体动力学分析系统、颗粒动力学分析系统等多套岩土工程及爆炸冲击领域的数值仿真软件,上述软件在模拟多场耦合下岩土材料渐进破坏过程方面具有明显优势。
    本文主要承担工作为:基金支持、模型建立和算法指导。
    FENG Chun is a senior engineer at the Institute of Mechanics, Chinese Academy of Sciences, he has led the development of multiple numerical simulation software systems in the fields of geotechnical engineering and explosion impact over the past 10 years, including the CDEM solver, block dynamics analysis system, and particle dynamics analysis system. These software systems have significant advantages in simulating the progressive failure process of geotechnical materials under multi-field coupling.
    In this study, he is primarily responsible for funding support, model establishment, and algorithm guidance.
    E-mail: fengchun@imech.ac.cn
  • 基金资助:
    光合基金A类(202302017772);国家自然科学基金(52178324)

Distributed Computing Strategy for Three-Dimensional Continuous-Discontinuous Element Method

HUANG Junguang1,2(),FENG Chun2,*(),ZHANG Yiming3   

  1. 1. School of Civil Engineering and Transportation, Hebei University of Technology, Tianjin 300401, China
    2. Institute of Mechanics, Chinese Academy of Sciences, Beijing 100190, China
    3. School of Architecture Engineering, Zhejiang Sci-Tech University, Hangzhou, Zhejiang 310018, China
  • Received:2025-03-19 Online:2025-10-20 Published:2025-10-23
  • Contact: FENG Chun

摘要:

【目的】 建立使用岩土工程大规模复杂模型的分布式计算范式具有重要的意义。【方法】 采用ParMETIS图划分算法,根据网格节点图的拓扑关系生成网格节点区域分解方案。通过引入ghost单元和ghost接触,基于网格节点分区结果,实现连续-非连续计算任务的多子域数据集划分和计算集群的负载均衡。利用MPI非阻塞通信和内存映射共享技术,实现ghost节点信息的高效通信。最终,在计算集群中实现大规模三维连续-非连续问题的数据分布式计算。【结果】 在单节点或者多节点负载固定的情况下,提出的分布式计算策略能够实现连续-非连续单元法的弱扩展性,突破了单机百万单元的限制,成功满足了千万级别四面体单元的复杂计算需求。【结论】 该分布式计算策略有效提高了计算能力和扩展性,为连续-非连续单元法在大规模岩土工程问题的应用提供了可行的解决方案。

关键词: 连续-非连续单元法(CDEM), 分布式计算, 图划分, 负载均衡

Abstract:

[Objective] Establishing a distributed computing paradigm for large-scale complex geotechnical engineering models is of significant importance. [Methods] The ParMETIS graph partitioning algorithm is used to generate a grid node domain decomposition scheme based on the topological relationships of the grid node graph. By introducing ghost elements and ghost contacts, a multi-subdomain data partitioning for continuous-discontinuous computational tasks and load balancing of the computing cluster is achieved based on the grid node partitioning results. Using MPI non-blocking communication and memory-mapped sharing technology, efficient communication of the ghost node information is realized. Finally, large-scale three-dimensional continuous-discontinuous problems are solved through distributed computing on the computing cluster. [Results] Under conditions where the load is fixed either on a single node or across multiple nodes, the proposed distributed computing strategy achieves weak scalability for the CDEM, overcoming the limitations of a single machine with millions of elements, and successfully meeting the complex computational demands of tens of millions of tetrahedral elements. [Conclusions] This distributed computing strategy effectively enhances computing capability and scalability, providing a feasible solution for the application of the continuous-discontinuous element method in large-scale geotechnical engineering problems.

Key words: continuous-discontinuous element method (CDEM), distributed computing, graph partitioning, load balancing