AN INTEGRATED SOLUTION STRATEGY ON PARALLEL COMPUTERS

This paper deals with an integrated solution strategy on a distributed-memory parallel computer. There seems to be the need for such an approach to problem solving on parallel computers. The various steps are outlined for problem solving including discretization of the spatial domain, load partitioning and mapping onto the processor space and utilizing the same kind of code as that for serial computers. The main issues in parallel computing are load balancing and user-friendliness. These can be addressed here by specifying the domain on which the equation is to be solved, the processor topology and the equation to be solved together with boundary conditions. The programming modules required to solve the problem are essentially the existing sequential codes modified to include appropriate communication routines to exchange information at the interprocessor boundaries. The strategy, though general, has been illustrated by solving partial differential equations. A Poisson equation on an arbitrary domain is solved as a test problem