City Research Online

Abstract

This thesis addresses the problem of specifying, designing and implementing parallel computer architectures based on wafer scale integration (WSI). The requirements and constraints of WSI are considered and the class of computer architecture that is most suited to the technology is identified. This takes the form of a regular array of similar processors connected by a general purpose communications network. The communications function of the array is considered separately from the processing function.

Three routing algorithms for regular two dimensional arrays of processors are proposed. These are specified, and their properties are analysed. The performance of each of these is measured by simulation under varying conditions.

The problem of specifying and designing the processors is addressed next. A functional language engine is chosen as the target architecture. The processor specified and designed is a parallel graph reduction machine (named Cobweb) that uses directors as the instruction set. The programs executed on the machine are compiled from strictness analysed Hope+ via FLIC to a director and parallelism annotated directed acyclic program graph.

A specification of a single processor, using a novel object-oriented parallel graph rewrite notation (named Paragon) is given. A methodology for translating Paragon specifications into a hardware design is given. This methodology is applied to the Cobweb specification. The resulting design is seen to be inefficient, so the specification is transformed, whilst retaining its semantics, to make it more efficient, and the translation process applied again. The resulting design has been simulated and some of the results from the simulator are shown.

The COBWEB specification is expanded to a multiprocessor one. Some of the problems in producing a specification for this type of machine are discussed. This specification is used to produce a design.

The results from a simulation of the multiprocessor COBWEB along with the results from the communications network chapter are used to predict the performance of a multiprocessor WSI graph reduction machine.

The thesis ends with a discussion of the merits and problems of specification and evaluation of this type of computer architecture. The communications architecture is found to be especially suitable for WSI; the specification and design tools are found to be sufficiently powerful, although limited in their scope. Finally the conclusion is drawn, with caveats, that WSI is a suitable technology for parallel graph reduction.

Publication Type:	Thesis (Doctoral)
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science T Technology > T Technology (General)
Departments:	Doctoral Theses School of Science & Technology > School of Science & Technology Doctoral Theses School of Science & Technology > Computer Science

Export

Downloads