Solving the Navier-Stokes Equations on a Workstation Cluster
by
Stuart Norris
University of Sydney
Coauthors: Steven Armfield

Transient natural convection can be effectively modelled using finite volume Navier-Stokes solvers. However capturing the fine detail of the flow structures places a considerable demand on computational time and memory size. This demand can be met by using supercomputers, but a low cost alternative is to use a cluster of conventional workstations. An existing unsteady three-dimensional finite volume Navier-Stokes solver has been parallelised to demonstrate the viability of using a workstation cluster to model such flows.

Parallel programs for distributed memory machines can be written using either data parallel languages, such as High Performance Fortran (HPF), or with message passing libraries such as PVM and MPI. In order to select which method to use, the Conjugate Gradient linear solver from the Navier-Stokes code has been implemented using HPF, PVM and MPI. The solver codes are compared in terms of programming complexity and runtime speed. Run times and results will also be presented for the full Navier-Stokes code, parallelised with HPF, applied to a thermal intrusion flow.

Date received: August 4, 1999