Year: 1986
Type: dissertation
Abstract: This thesis mainly covers the design and analysis of asynchronous\nparallel algorithms that can be run on MIMD (Multiple Instruction\nMultiple Data) parallel computers, in particular the NEPTUNE system at\nLoughborough University. Initially the fundamentals of parallel computer\narchitectures are introduced with different parallel architectures being\ndescribed and compared. The principles of parallel programming and the\ndesign of parallel algorithms are also outlined. Also the main\ncharacteristics of the 4 processor MIMD NEPTUNE system are presented,\nand performance indicators, i.e. the speed-up and the efficiency factors\nare defined for the measurement of parallelism in a given system.\nBoth numerical and non-numerical algorithms are covered in the\nthesis. In the numerical solution of partial differential equations,\na new parallel 9-point block iterative method is developed. Here, the\norganization of the blocks is done in such a way that each process\ncontains its own group of 9 points on the network, therefore, they can\nbe run in parallel. The parallel implementation of both 9-point and 4-\npoint block iterative methods were programmed using natural and redblack\nordering with synchronous and asynchronous approaches. The\nresults obtained for these different implementations were compared and\nanalysed.\nNext the parallel version of the A.G.E. (Alternating Group Explicit)\nmethod is developed in which the explicit nature of the difference\nequation is revealed and exploited when applied to derive the solution\nof both linear and non-linear 2-point boundary value problems. Two\nstrategies have been used in the implementation of the parallel A.G.E.\nmethod using the synchronous and asynchronous approaches. The results\nfrom these implementations were compared. Also for comparison reasons\nthe results obtained from the parallel A.G.E. were compared with the ~\ncorresponding results obtained from the parallel versions of the Jacobi,\nGauss-Seidel and S.O.R. methods. Finally, a computational complexity\nanalysis of the parallel A.G.E. algorithms is included.\nIn the area of non-numeric algorithms, the problems of sorting and\nsearching were studied. The sorting methods which were investigated\nwas the shell and the digit sort methods. with each method different\nparallel strategies and approaches were used and compared to find the\nbest results which can be obtained on the parallel machine.\nIn the searching methods, the sequential search algorithm in an\nunordered table and the binary search algorithms were investigated and\nimplemented in parallel with a presentation of the results. Finally,\na complexity analysis of these methods is presented.\nThe thesis concludes with a chapter summarizing the main results.
Author: Hilal M. Yousif
Language: English
Subfield: Hardware and Architecture
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W838958356