Computer Architecture and High Performance Computing, Symposium on (2009)
Sao Paolo, Brazil
Oct. 28, 2009 to Oct. 31, 2009
This study presents a new parallel Finite Element Method (FEM) strategy designed for coarse grain distributed memory systems. The adopted communication protocol is Message Passing Interface (MPI) and tests are carried out in a cluster of PCs. Compressed data structure is used to store the Hessian matrix in order to optimize memory usage and to use the parallel direct solver MUMPS. The new partitioning paradigm is based on structural finite element nodes, not elements (as usually done in references), resulting in an overlapping algorithm, where a reduced amount of information should be allocated and manipulated to integrate finite elements. The main advantage of the nodal partitioning is the performance improvement of the Hessian matrix assembly and the natural ordering to improve the system solution. Numerical examples are shown in order to demonstrate the efficiency and scalability of the proposed algorithm.
parallel computing, message passing interface, finite element method, coarse grain simulation
C. L. Kawabata, M. K. Neto, H. B. Coda and W. S. Venturini, "Finite Element Nodal Approach for Parallel Processing of Non Linear Shell Analysis," Computer Architecture and High Performance Computing, Symposium on(SBAC-PAD), Sao Paolo, Brazil, 2009, pp. 109-116.