A taxonomy of parallel computer architectures

Discusses a framework within which it is possible to compare the overall performance of a wide variety of parallel computing systems. In this context, a computing system comprises components at several levels of abstraction, ranging from applications, through algorithms and various levels of language, to hardware implementation. The basis of the desired framework will be a series of classification schemes or taxonomies, at the various levels of abstraction, together with a set of mapping strategies that guide selection of appropriate options at each level, top-down. At present, it is not feasible to address all levels simultaneously, and so the author concentrates on the `lowest´ level of abstraction (i.e. the hardware implementation)