J.W. Harley
University of Newcastle upon Tyne. 1995.
In the 1980s, multiple-processor computers (multiprocessors) based on conventional processing elements emerged as a popular solution to the continuing demand for ever-greater computing power. These machines offer a general-purpose parallel processing platform on which the size of program units which can be efficiently executed in parallel - the "grain size" - is smaller than that offered by distributed computing environments, though greater than that of some more specialised architectures. However, programming to exploit this medium-grained parallelism remains difficult. Concurrent execution is inherently complex, yet there is a lack of programming tools to support parallel programming activities such as program design, implementation, debugging, performance tuning and so on. In helping to manage complexity in sequential programming, visual tools have often been used to great effect, which suggests one approach towards the goal of making parallel programming less difficult. This thesis examines the possibilities which the dataflow paradigm has to offer as the basis for a set of visual parallel programming tools, and presents a dataflow notation designed as a framework for medium-grained parallel programming. The implementation of this notation as a programming language is discussed, and its suitability for the medium-grained level is examined.