Cluster oriented architecture for the mapping of parallel processor networks to high performance applications

Publisher Website
Google Scholar
Abstract
Massively parallel computer architectures have been discussed since the very beginning of electronic computing. The major design goal was clearly getting a superior system performance. Today, with supercomputers, most commonly used, this has changed to match the computing power of supercomputers for a fraction of their costs. Up to now most parallel computer systems have been build up by university institutes for research work with few commercial success.The recent progress in computer architecture design has made parallel processor networks much more feasible. Due to combinatorical problems in processor interconnection networks, cluster oriented approaches are the most favorable way for designing parallel systems. This requires a special interconnection hardware as well as mapping algorithms.In this paper we will present our new parallel cluster oriented computer system based on the INMOS transputer technology. This system includes a special network manager supported by dedicated hardware to make use of advanced dynamic mapping strategies in an efficient and flexible way.

This publication has 0 references indexed in Scilit: