Date of Award

Winter 1-31-1994

Document Type

Thesis

Degree Name

Master of Science in Electrical Engineering - (M.S.)

Department

Electrical and Computer Engineering

First Advisor

Sotirios Ziavras

Second Advisor

Denis L. Blackmore

Third Advisor

John D. Carpinelli

Abstract

The choice of a topology for the interconnection of resources in a distributed-memory parallel computing system is a major design decision. The direct binary hypercube has been widely used for this purpose due to its low diameter and its ability to efficiently emulate other important structures. The aforementioned strong properties of the hypercube come at the cost of high VLSI complexity due to the increase in the number of communication ports and channels per node with an increase in the total number of nodes. The reduced hypercube (RH) topology, which is obtained by a uniform reduction in the number of links for each hypercube node, yields lower complexity interconnection networks compared to hypercubes with the same number of nodes, thus permitting the construction of larger parallel systems. Furthermore, it has been shown that the RH at a lower cost achieves performance comparable to that of a regular hypercube with the same number of nodes. A very important issue for the viability of the RH is to investigate the efficiency of embedding frequently used topologies into it. This thesis proposes embedding algorithms for three very important topologies, namely the ring, the torus and the binary tree. The performance of the proposed algorithms is analyzed and compared to that of equivalent embedding algorithms for the regular hypercube. It is shown that these topologies are emulated efficiently on the RH. Additionally, two already proposed routing algorithms for the RH are evaluated through simulation results.

Share

COinS