|
| Research on optimization techniques for parallel finite element analysis based on local communication operations |
| Received:March 16, 2020 Revised:July 08, 2020 |
| View Full Text View/Add Comment Download reader |
| DOI：10.7511/jslx20200316001 |
| KeyWord:finite element stiffness matrix sparse matrix parallel algorithm concrete specimen |
| Author | Institution |
| 吴建平 |
国防科技大学 气象海洋学院, 长沙 |
| 蒋涛 |
国防科技大学 气象海洋学院, 长沙 ;国防科技大学 计算机学院, 长沙 |
| 彭军 |
国防科技大学 气象海洋学院, 长沙 |
| 银福康 |
国防科技大学 气象海洋学院, 长沙 |
| 杨锦辉 |
国防科技大学 气象海洋学院, 长沙 |
|
| Hits: 984 |
| Download times: 622 |
| Abstract: |
| In this paper,for finite element analyses,based on the existing simple parallel algorithm with global communication operations,the core algorithms are optimized with sparse data structures and local communication operators.These strategies reduce the number of processors and traffic involved in communication.At the same time,by using non-blocking communication operations and executing the communication-free computations first,the overlapping of communication with computations is carried out to effectively hide the communication overhead.The experimental results show that the optimized algorithm has been greatly improved,especially for the multiplication of a sparse matrix by a dense vector and the assembly of elemental contributions.In addition,with the increase of the number of tasks,the improvement is more and more significant. |
|
|
|