摘要
For node-based local mesh generation (NLMG), existing parallel algorithms are not only unsuitable but also poor in efficiency. So we propose a new and what we believe to be an efficient NLMG parallel algorithm. In the paper, we explain a new parallel algorithm in detail. We propose three topics of our explanation; (A) CDP (Constrained Delaunay Path)-based domain partition algorithm; (B) NLMG parallel algorithm; (C) further discussion of NLMG parallel algorithm. Using the CDP to partition an arbitrary node-given domain, we can eliminate the frequent communications among processors in the local mesh generation process and decrease the communications to the fullest degree during the stiffness matrix computing. We point out that the parallel algorithm naturally implements the seamless link between mesh generation and stiffness matrix computing, which fully demonstrates the potential efficiency of the node-based local finite element method. We discuss the great effect of load balance on parallel efficiency and discuss also the appropriate numbers of subdomains and processors. Numerical experimental results show preliminarily that the NLMG parallel algorithm is effective and reliable, especially for large-scale problems; and if the number of chosen processors is appropriate, the parallel efficiency can reach nearly 90%.
| 源语言 | 英语 |
|---|---|
| 页(从-至) | 731-735 |
| 页数 | 5 |
| 期刊 | Xibei Gongye Daxue Xuebao/Journal of Northwestern Polytechnical University |
| 卷 | 24 |
| 期 | 6 |
| 出版状态 | 已出版 - 12月 2006 |
学术指纹
探究 'New and efficient node-based local mesh generation parallel algorithm' 的科研主题。它们共同构成独一无二的学术指纹。引用此
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver