Numerical Simulation of Complex Fluid Flows on MIMD Computers

The paper analyses the efficiency of parallel computation of incompressible fluid flows using a fully implicit finite volume multigrid method. The parallelization is achieved via domain decomposition, which is chosen for its suitability in complex geometries when blockstructured or unstrucutured grids are employed. Numerical efficiency (increase of computing effort to reach converged solution) and parallel efficiency (increase of runtime due to local and global communication) are analysed for a typical recirculating flow induced by buoyancy. Good efficiencies are found and the possibilities for further improvement by avoiding some global communication or by simultaneous computation and communication are indicated.