Parallelism in production systems

摘要: Production system programs, on the surface, appear to be capable of using large amounts parallelism--it is possible match each production in a program data memory parallel. The thesis shows that practice, however, speed-up obtainable from parallelism quite limited, around 15-fold as compared initial expectations 100-fold 1000-fold. main reasons for limited are: (1) there are only small number productions require significant processing result change working memory; and (2) variation requirement these productions. To obtain fraction available, proposes parallel version Rete algorithm which exploits at very fine grain. It further suggests suitable architecture exploit fine-grained shared-memory multiprocessor, with 32-64 high performance processors. For scheduling grained tasks (consisting about 50-100 instructions), hardware task scheduler proposed. The presents set simulation results systems exploiting different sources parallelism. points out features existing programs limit solutions some bottlenecks. show suggested multiprocessor (with individual processors performing 2 MIPS), it execution speeds 5000-27000 element changes per second. This corresponds 5-fold over best known sequential implementation MIPS processor. also higher than obtained by other proposed implementations systems.