Efficient algorithms for the optical multi-trees (OMULT) architecture.
Date of Award
CC BY-NC-ND 4.0
In this thesis, we have reported our investigations on efficiently implementing algorithms on the recently proposed Optical Multi-Trees (OMULT) multi-processors interconnection architecture that uses both electronic and optical links among processors. We have investigated algorithms for matrix multiplication of two matrices of size n2 x n2 and two matrices of arbitrary size, the prefix-sum of a series and some fundamental computational geometry problems. We show that some common algorithms for computational geometry---finding the convex hull, the smallest enclosing box, the empirical cumulative distribution function and the all-nearest neighbor problems of n data points can be computed on the OMULT network in O(log n) time, compared to O(√n) algorithms on the Optical Transpose Interconnection System (OTIS) mesh for each of these problems. Finally we have implemented our algorithm for matrix multiplication using the SimJava simulation tool and feel that this is a convenient environment for testing such parallel algorithms.Dept. of Computer Science. Paper copy at Leddy Library: Theses & Major Papers - Basement, West Bldg. / Call Number: Thesis2004 .I85. Source: Masters Abstracts International, Volume: 43-05, page: 1751. Adviser: Subir Bandyopadhyay. Thesis (M.Sc.)--University of Windsor (Canada), 2004.
Islam, Mohammad Rabiul., "Efficient algorithms for the optical multi-trees (OMULT) architecture." (2004). Electronic Theses and Dissertations. 1879.