A Knowledge-Based Scientific Parallel Programming Environment
Karsten M. Decker, Jiri J. Dvorak, René M. Rehmann · Birkhäuser Basel eBooks · 1994
Distributed memory parallel systems are still lacking simple and economic programmability, since, from the scientific programmer’s point of view, first generation tools and environments for distributed memory parallel processor systems are largely insufficient. In this paper, we present an overview of our approach to parallel distributed programming environments overcoming these limitations: a knowledge-based problem-solving environment for programming parallel distributed systems. The environment features a problem-oriented specification formalism and is based on a skeleton-oriented programming methodology. The large set of fine-grain algorithmic skeletons used in the system is managed with a knowledge-based component. Skeletons are completed with computational components by means of automatic program synthesis techniques. We describe in detail a working prototype that, for the class of stencil-based problems, supports transparent programming of parallel distributed systems, while successfully addressing and solving some of the problems of performance, portability and software reuse. These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.