A characterization of parallel systems

a taxonomy for parallel processing systems is presented which has some advantages over previous taxonomies. The taxonomy characterizes parallel processing systems using four parameters: topology, communication, granularity, and operation. These parameters and used repetitively in a hierarchical fashion to produce a taxonomic structure which is extensible to the level of detail desired. Topology describes the structure of the priniciple interconnections. Communication describes the flow of data and programs through the system. Granularity describes the size of the largest repeated element, or grain. Operation describes the important functional properties of each grain, especially the ratio of storage to logic circuitry. Granularity and topology are structural parameters, while operation and communication are functional parameters which describe the behavior of the system components. A final section of this paper includes examples of the application of the taxonomy to several parallel processing systems.