193	1997	Peter Olsson	peter@tdb.uu.se	Object-Oriented Software Tools for Parallel Computing with Distributed Memory

                               Abstract

Software tools have been developed for simplifying the use of parallel 
computers with distributed memory in the context of scientific computing. 
An object-oriented approach has been used, leading to tools that are easy 
to maintain, modify, and extend. The tools are intended for time-dependent 
partial differential equations solved by using explicit finite difference 
methods.

Solvers based on these tools are portable over a wide range of computers and 
message-passing libraries. The experimental program used in this report has 
been executed (without changes) on the following platforms: SUN workstations 
in an Ethernet network with PVM, Digital AlphaServer 8200 with PVM and MPI, 
IBM SP2 with MPL, Intel Paragon XP with NX, and Parsytec GC/Powerplus with 
PARIX.

The experimental results show that the tools give good parallel performance 
and that the extra software layer introduced gives virtually no degrading of 
the performance. The results also show that the tools can be used as a basis 
for program development.