Abstract
To shorten calculation time in finite element simulation by using multithreading computer, a parallel algorithm for stiffness matrix decomposition based on threadpool method is proposed. Firstly, a decomposition method of applicability to parallel computation is put forward by transferring the Cholesky's LLT method. Then, the threadplool is employed to generate multithreading for repeating use and the optimization is conducted considering load-balancing of each thread. Finally, numerical tests by using proposed algorithm in decomposition of one-dimensional array stored stiffness matrix are carried out on different calculation platforms with multi-processors. It is shown that the parallel algorithm can overcome the limitations of OpenMP when being applied in nested loops and is of high efficiency on stiffness matrix decomposition with low platform demands. The algorithm has explicit concept and minor programming difficulty and is applicable to solve problems caused by limitation of OpenMP in particular.
Publisher
Trans Tech Publications, Ltd.
Reference11 articles.
1. Heejo Lee, Jong Kim, Sung Je Hong, Sunggu Lee. Task scheduling using a block dependency DAG for block-oriented sparse Cholesky decomposition. Parallel Computing 29(2003), 135–159.
2. Zhang Jianfei, Jiang Hongdao. Preprocessing of FEM parallel computation. Chinese Journal of Computational Mechanics 20(2003), 500-503.
3. LI Hai-jiang, YANG Gang, YI Nan-gai. Object-oriented serial/ Parallel finite element analysis system. Chinese Journal of Computational Mechanics 23(2006), 500-503.
4. D. Zheng, T. Y. P. Chang. Parallel cholesky method on MIMD with shared memory. Computers & Structure 56(1995), 25-38.
5. William Gropp, Ewing Lusk, Anthony. Using MPI - 2nd Edition: Portable Parallel Programming with the Message Passing Interface , The MIT Press, (1999).