Date of Award

1997

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computer Science

First Advisor

Si-Qing Zheng

Abstract

In this doctoral research, we propose several novel protocols and topologies for the interconnection of massively parallel processors. These new technologies achieve considerable improvements in system performance and structure simplicity. Currently, synchronous protocols are used in optical TDM buses. The major disadvantage of a synchronous protocol is the waste of packet slots. To offset this inherent drawback of synchronous TDM, a pipelined asynchronous TDM optical bus is proposed. The simulation results show that the performance of the proposed bus is significantly better than that of known pipelined synchronous TDM optical buses. Practically, the computation power of the plain TDM protocol is limited. Various extensions must be added to the system. In this research, a new pipelined optical TDM bus for implementing a linear array parallel computer architecture is proposed. The switches on the receiving segment of the bus can be dynamically controlled, which make the system highly reconfigurable. To build large and scalable systems, we need new network architectures that are suitable for optical interconnections. A new kind of reconfigurable bus called segmented bus is introduced to achieve reduced structure simplicity and increased concurrency. We show that parallel architectures based on segmented buses are versatile by showing that it can simulate parallel communication patterns supported by a wide variety of networks with small slowdown factors. New kinds of interconnection networks, the hypernetworks, have been proposed recently. Compared with point-to-point networks, they allow for increased resource-sharing and communication bandwidth utilization, and they are especially suitable for optical interconnects. One way to derive a hypernetwork is by finding the dual of a point-to-point network. Hypercube $Q\sb{n},$ where n is the dimension, is a very popular point-to-point network. It is interesting to construct hypernetworks from the dual $Q\sbsp{n}{*}$ of hypercube of $Q\sb{n}.$ In this research, the properties of $Q\sbsp{n}{*}$ are investigated and a set of fundamental data communication algorithms for $Q\sbsp{n}{*}$ are presented. The results indicate that the $Q\sbsp{n}{*}$ hypernetwork is a useful and promising interconnection structure for high-performance parallel and distributed computing systems.

ISBN

9780591723984

Pages

113

Share

COinS