| Summary: | The dissertation investigates a data flow programming style for developing efficient and machine-independent scientific programs on multiprocessor computers. I have designed and implemented a programming system called Neptune. Both functional and data parallelism are incorporated into Neptune. A coarse-grain data flow model is used to explicitly specify functional parallelism. Data parallelism is represented within the data flow model by an activity decomposition model which ensures efficient execution of data parallel computation. Four scientific applications have been implemented in this data flow style to evaluate execution performance. === The machine-independence of the data flow model is demonstrated by obtaining speedup performance on three different parallel architectures (Sun network, Sequent Balance 12000 and a Cray Y-MP/464). The dissertation provides a detailed description of the implementation of the runtime environment on three main classes of parallel architectures (network, shared memory and distributed memory). Each implementation takes advantage of an architecture and provides a suitable data mapping strategy for minimizing overhead and optimizing resource utilization. === Developing parallel programs presents a major difficulty. In addition to the sequential problems, the programmer has to design and debug parallel constructs which express concurrency. The dissertation describes the Neptune programming system, which is specially designed for supporting a data flow methodology. Neptune provides an effective visual environment for designing and debugging data flow programs. Applications developed and debugged on a network of workstations can be scaled up and run on multiprocessor computers without requiring any software modifications. === Source: Dissertation Abstracts International, Volume: 51-12, Section: B, page: 5982. === Major Professor: Gregory A. Riccardi. === Thesis (Ph.D.)--The Florida State University, 1990.
|
|---|
