Multimicroprocessor resource allocation

This thesis investigates the problems of allocating the data and code address spaces of a concurrent program onto the stores of a given multiprocessor computer architecture, and the allocation of the processes of the program to the processors of the architecture. The minimum required of this resource allocation is to produce a legal mapping of the resources onto the multiprocessor computer. It will also attempt to give the most efficient mapping, and allow the user to guide this activity. This thesis describes the methods developed to implement this, which includes the specification of the structures of both the program and the computer architecture in a machine understandable form, and the design of algorithms to perform the allocation. With the resulting techniques the emphasis is upon small scale multiprocessor computer architectures running dedicated concurrent programs. The resource allocation scheme results in a fixed allocation of the parts of a single program to a possibly nonstandard and specially tailored multiprocessor architecture. This would find little application with large regular mainframe multiprocessor computers executing time shared operating system programs, where the allocation of resources is highly dynamic and unknown at compile time.