Initial Assessment of Architectures for Production Systems

Abstract

Although production systems are appropriate for many applications in the artificial intelligence and expert systems areas, there are applications for which they are not fast enough to be used. If they are to be used for very large problems with severe time constraints, speed increases are essential. Recognizing that substantial further increases are not likely to be achieved through software techniques, the PSM project has begun investigating the use of hardware support for production system interpreters. The first task undertaken in the project was to attempt to understand the space of architectural possibilities and the trade-offs involved. This article presents the initial findings of the project. Briefly, the preliminary results indicate that the most attractive architecture for production systems is a machine containing a small number of very simple and very fast processors.