John R. Rice
Department of Computer Science
Purdue University
W. Lafayette, IN 47907, USA

A brief overview of the goals and structures of problem solving environments (PSEs) is given. This is followed by a more focused discussion of the problems of parallel computing in PSEs. The current methodologies for parallel computing are too complex and too specialized to be accessible for most engineers and scientists. Thus a PSE for scientific computing must hide most or all of the considerations of load balancing, parallelization of algorithms, distributed memory, message passing, etc. The techniques for hiding or automating these methodologies are reviewed, along with their current and future potential for suc- cess. A second approach for parallel computing in scientific PSEs is to use an intuitively understandable paradigm based on collaborating software systems and software agents. This approach is still untested and it still requires that many spe- cialized techniques of parallelism be hidden by the PSEs. But it might provide a natural and effective way for users to describe or specify the global parallelism in applications. If this is successful then the automation of the lower level parallelization tasks is much more feasible.
( slides )