结构模态多级分层并行计算方法  

A multilevel hierarchical parallel algorithm for finite element structural modal analysis

作　　者：喻高远 楼云锋 李俊杰 金先龙[1] YU Gaoyuan;LOU Yunfeng;LI Junjie;JIN Xianlong(School of Mechanical Engineering,Shanghai Jiao Tong University,Shanghai 200240,China;Aerospace System Engineering Shanghai,Shanghai 201108,China)

机构地区：[1]上海交通大学机械与动力工程学院,上海200240 [2]上海宇航系统工程研究所,上海201108

出　　处：《振动与冲击》2023年第16期19-25,共7页Journal of Vibration and Shock

基　　金：国家自然科学基金(11772192)。

摘　　要：基于稀疏存储技术和传统并行模态综合法提出了一种有限元结构模态分析多级分层并行计算方法。该方法在两级分区4次变换策略的基础上不仅实现了大量数据的分布式稀疏存储,提高了数据的内存访问效率,而且实现了系统整体缩减后的广义特征方程规模的有效降低,大幅度减少了广义特征方程的求解时间。此外,它还利用计算任务和异构众核集群硬件体系结构映射实现了计算过程的多级并行,不仅有效改善了不同层级的负载均衡,而且通过通信分离有效提高了通信效率。因此,它能够充分利用异构众核分布式存储并行计算机的体系结构特点提升大规模有限元模态并行计算效率。数值算例表明,相比于传统的并行模态综合法,稀疏存储格式模态多级分层并行方法能够大幅度节省内存空间和提高计算效率。A multilevel hierarchical parallel modal synthesis algorithm for large-scale finite element modal analysis was proposed based on the sparse storage techniques and parallel conventional modal synthesis algorithm.Based on two-level partitioning and quartic transformation strategies,the proposed method not only improves memory access rate through the distributed sparse storage of a large amount of data,but also reduces the solution time of the interface equation through the reduction in interface equations’size.Moreover,a multilevel hierarchical parallelization of the computational procedure was introduced to enable the separation of the communication of inter-nodes,heterogeneous-core-groups,and inside-heterogeneous-core-groups through mapping computing tasks to various hardware layers,which can not only efficiently achieve the load balancing at different layers,but also significantly improve communication rate through the hierarchical communication.Hence,it can improve the efficiency rates of parallel computing of large-scale finite element modal analysis by fully exploiting the architecture characteristics of heterogeneous multicore clusters.Test results show that,compared with the parallel conventional modal synthesis algorithm,the proposed method with sparse storage format can considerably save memory space and significantly improve computational efficiency.

关 键 词：异构众核 多级分层并行 负载均衡 稀疏存储 大规模模态分析 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论] O246.6[自动化与计算机技术—计算机科学与技术]