Hierarchical multiprocessor organizations

Abstract

The development of LSI components has induced a substantial research effort into the possibility of applying conglomerations of small processors on some computational problems previously limited to large processors. The primary advantages of “multi-microprocessor architectures” are their potential for improvements in cost, reliability, and possibly speed over conventional large computers. However, the realization of these advantages requires improvements in the techniques necessary to translate problems into parallel algorithms for multi-microprocessors, and in the architectures of the multi-microprocessors themselves. This paper describes an architecture currently being investigated at SUNY - Stony Brook. The first section discusses various strategies of design. The second section describes the structure of the proposed system. Finally, the third section discusses some classes of problems for which the architecture is suited.