This paper presents a new taxonomy for computer architectures that is based on partitioning the architectures into functional units and considering the information flow between these units. This taxonomy forms a two-level structure. At the higher level, we have abstract machines. This level extends Flynn’s classification of computer architectures to capture the significant differences between different types of parallel machine. Skillicorn classifies architectures by the number of instruction processors (i.e., functional units which interpret instructions), the number of data processors (i.e., functional units which transform data), the number of memory hierarchies, and the type of interconnection between these functional units. At the lower level, we have machine implementations. This level is based on a state diagram view of the processors. It allows the user to specify the degree of pipelining, for instance, and to discriminate between pipelined and non-pipelined processors.

The paper contains a number of case analyses, which include von Neumann, graph reduction, and dataflow architectures. The author analyzes an example of a loosely coupled graph reduction system, the Flagship system, in particular depth to demonstrate the taxonomy’s ability to cover unconventional machines.

The paper also contains a brief but adequate survey of a number of classification schemes proposed by other authors and presents a set of somewhat debatable arguments in favor of classifications of this type. Of course, any taxonomy can only capture a subset of the set of parameters that characterize the architecture of a computer system. The taxonomy proposed in this paper concentrates mainly on physical organization and hardware functionalities, leaving out certain important aspects such as architectural supports for operating systems and high-level language implementation. The author seems to be aware of this inherent limitation that reduces the advantages of a practical application of his taxonomy in the design of a new architecture. On the other hand, this well-written paper should be useful reading for the nonspecialist, who will find a clear identification of a number of important aspects of computer architecture and organization.