In this paper, a new architecture for the parallel execution of production systems called MANJI is proposed and evaluated. MANJI consists of tens of powerful processing units (PUs). All the PUs have special shared memory (Paged Multiple Readout Memory, abbreviated P-MRM), and communicate through special memory architecture (Paged Receiver Selectable Multicast, abbreviated P-RSM). The working memory is stored in P-MRM, and shared by all PUs. OPS5 is adopted for the specification language and is compiled into a data flow graph. This graph is divided and distributed to each PU. According to this graph, the PUs execute the production system in parallel and communicate with other PUs through P-RSM if necessary. The communication capacity of these two memory architectures is evaluated.
|Number of pages
|Proceedings of the Hawaii International Conference on System Science
|Published - 1987 Jan 1
ASJC Scopus subject areas
- General Computer Science