A Comparative Study of Data-Flow Architectures

educational or research purposes on condition that (1) this copyright notice is included, (2) proper attribution to the author or authors is made and (3) no commercial gain is involved. Abstract Partly as a result of the diversity of the architectures and system software, Data-Flow systems have not been compared directly. In this paper, three widely different Data-Flow systems are compared using a relatively uniform metric which is representative of the actual amount "work" performed by these systems to execute a small collection of common benchmarks. One feature, common to all Data-Flow systems, forms the basis for a metric for "work", the creation of tokens. By counting the creation of tokens in all parts of a Data-Flow system, an accurate measure of the work performed can be obtained. The three systems compared are the Manchester Data-Flow Machine, the Stateless Data-Flow Architecture (also from Manchester), and the CSIRAC II machine. All machines perform approximately the same amount of work when solving a collection of benchmark problems, but two systems (CSIRAC II and SDFA) exhibit successful exploitation of a memory hierarchy.