Transformations of Concurrent Algorithms for Highly Parallel Systems: A One Year Project Summary Report.
Abstract
It has become a certainty that Multiple Instruction Stream, Multiple Data Stream (MIMD) parallel architectures are going to play a major role in all aspects of high speed computer design for the foreseeable future. What is not clear is whether we will be able to devise a means to design algorithms and software for these machines that transcends our current ad-hoc, nonportable techniques. In this research project the author has focused on the portability issue from the perspective of parallel algorithm design and how it effects the internal organization of advanced compilers. The eventual goal of the project is to produce an expert system that can help users transform large, complex applications from one highly parallel machine to another. The basic strategy has been to follow the following plan: 1) Build an experimental research laboratory for parallel computation; 2) Design and experimental program of research that would shed light on the problems involved with restructuring parallel programs for different machines; 3) Attempt to provide a mathematical characterization of the properties of the machines and how algorithms must be restructured to run on them; and 4) Attempt to design a model of machine architecture that can be embedded into the inference engine or knowledge base of an expert system for program restructuring.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 01, 1987
- Accession Number
- ADA190236
Entities
People
- Dennis Gannon
Organizations
- Indiana University Bloomington