Author

Prabhala

Date of Award

4-1992

Degree Name

Master of Science

Department

Computer Science

First Advisor

Dr. Naveed A. Sherwani

Second Advisor

Dr. Alfred Boals

Third Advisor

Dr. Ajay Gupta

Access Setting

Masters Thesis-Open Access

Abstract

Networked multiprocessing architectures for parallel com putation offer an alternative to high cost supercomputing. Recently hypercube has emerged as the most versatile architecture for parallel com putations. However, the number of nodes m in a hypercube is a power of 2, 2d, where d is the dimension of the hypercube. In practice, it m ay not be possible to have a complete hypercube because the cost of upgradation is proportional to the num ber of nodes. Incomplete and Composite hypercubes help remove the exponential node(and hence cost) constraint.

In this thesis we establish the equality of m-node composite and incomplete hypercubes. Then we identify a class of algorithms designated Fully Normal Algorithms for implementing a variety of algorithms on composite hypercubes. We define critical size of a composite hypercube to help identify the performance bounds and compute the speedup achieved. Finally, we develop a set of graph algorithms that can be used as building blocks.

Share

COinS