Parallel Finite Element Analysis of Incompressible Viscous Flow

概要

論文の詳細を見る
A parallel finite element method for incompressible viscous flows is presented. This method is based on the GSMAC (Generalized Simplified Marker And Cell) FEM in which simultaneous relaxation of velocities and pressures is employed. The computed field is decomposed into several domains. One processor is assigned to each domain. By using the method of domain decomposition and the calculation algorithm which is computed Node-by-Node, an algorithm for parallel calculations which needs only data communication of the velocities is proposed. Parallel computers used here are the NEC Cenju-3 and the Hitachi SR2201. The performance of this method is examined by analyses of the two dimensional lid-driven square cavity flows at the Reynolds number of 1000. As the results, high performance is achieved in the computing speed as well as in the parallel efficiency. The efficiency parameter of the parallel computations becomes about 1.0,when the number of the nodes assigned to one processor is greater than about 1600 on the NEC Cenju-3. The higher performance is obtained on the Hitachi SR2201 than on the NEC Cenju-3 in the present studies. When the ratio of the total number of nodes to the number of processors is greater than about 1300,the efficiency parameter becomes to about 1.0.
東海大学の論文