Multigrid efficiency for complex flow simulations on distributed memory machines

This paper investigates the extent to which a combined parallel and numerical efficiency can be maintained for the multigrid solution of problems on complex domains, using the incompressible Navier-Stokes equations on block-structured grids as a representative and non-trivial example. Algorithmic variants are introduced, which avoid a severe degradation of the convergence behaviour as the number of processors employed is increased. Numerical experiments are presented together with efficiency results for the Intel iPSC/860.