Cenju-3 parallel computer and its application to CFD

The exploitation of the effectiveness of parallel computers for computational fluid dynamics (CFD) requires an efficient parallel algorithm for numerical solution of the Navier-Stokes equations. This paper discusses the parallelization of an incompressible Navier-Stokes equation solver, and its implementation to the Cenju-3 parallel computer. We mainly focus our attention to the parallel solution of the discrete Poisson equations, which solution consumes the most of the computational time. Two parallel linear solvers are implemented based on a Rectangular Domain Decomposition Method (RDDM) and on the SMAC scheme. A parallel linear solver implemented is the Blocked MICCG (B-MICCG) method introduced by Washio and Hayami. The other linear solver is the Multi-Grid preconditioned Bi-CGSTAB(MG-Bi-CGSTAB) method. Numerical experiments has been conducted on Cenju-3 for up to 25 processors. Although the MG-Bi-CGSTAB method produces poorer speedup than the B-MICCG method, the MG-Bi-CGSTAB method is about 4 times faster than the B-MICCG method in terms of the total execution time