Abstract

One of the areas of computer architecture with the most rapid growth is that of parallel multiprocessors - an interconnection of anywhere from a few processing elements (PEs) to millions of PEs operating in parallel on the same problem. Attempts at solutions to many difficult problems in the past resulted in a rapid increase in complexity of software. With the continuing decline in the cost of hardware, attention has turned toward mapping such problems on massively parallel machines. For example, knowledge base searching can be done by pursuing many paths in parallel rather than following a complex algorithm that attempts to guide a serial search in an optimum fashion. The result is that many more computations are performed when following the parallel approach, with most not leading to a solution; however, the correct solution will most often be arrived at much sooner.

© 1988 Optical Society of America