Design of the matching unit of a massively parallel dataflow computing system

摘要

This paper describes a Massively Parallel Computing System, named Recursively Structured Parallel Computer-RSPC, designed to work as a dataflow distributed computer, and presents the design of the matching store of each node. That machine, structured as a multicomputer based on hierarchical buses, can have thousands of nodes. Each node, acting as a dataflow processor, can execute a part of a dataflow program, cooperating to solve the overall problem. Several strategies for the matching store implementation are discussed and the analysis of the techniques for highly efficient memory search is presented.