Overset-Grids Method for Flow Computation on Complex Geometry and its Performance Improvement with Hybrid Parallelization using Shared Memory and Distributed Memory Techniques
Please login to view abstract download link
The hybrid method for the overset-grids method which capable to compute flows of complex geometries combining the domain decomposition and parallelization methods with the share memory type is developed. An in-house structured CFD solver based on the overset-grid method is utilized. The governing equations are 3D RANS equations for the incompressible flow and spatial discretization is based on the finite-volume method. An interface capturing method based on the single phase level-set approach is employed to capture free surfaces. The weight values for the overset-grid interpolation is determined by the in-house system which is based on the Ferguson spline interpolation. The computational grids which include the overset information are divided for the distributed memory type method using the MPI based on the two ways, smoothing with the numbers of the solved cells or sending the flow variables. The present method is applied for the computations of the flow around the ship hull account for the ship motions with incoming waves and the propeller rotating propulsive condition using the multi-core processors and infiniband system. The hybrid method with the certain division number for the domain decomposition and the thread number achieves faster computation comparing with the flat MPI and pure OpenMP methods.