"Physical behavior of computing networks with an emphasis on routing" by Peter George Drexel

Date of Award

Winter 1993

Project Type

Dissertation

Program or Major

Engineering

Degree Name

Doctor of Philosophy

First Advisor

Andrzej Rucinski

Abstract

Combining powerful, local computing capability with inexpensive packet switched communication invites the creation of a new entity--the computing network. A computing network is a group of computer systems, interconnected by a packet switched network, which cooperate to accomplish a specific task.

Efficient communication between computer systems is critical. The accepted method of dealing with packet congestion is to limit network operation to linear regions ($\le$80 percent of total capacity).

Because of large parameter spaces, the behavior of computing networks is complex and not easily predictable with current techniques except in the simplest, linear region of operation. Since traditional approaches are inadequate, examining computing networks from a different perspective is desirable.

The complex behavior of a computing network in some sense emulates the behavior of many natural systems. The computing network may be thought of as a "pseudo-physical system" (PPS). A PPS is any complex, artificial system which exhibits a critical subset of the characteristics of a natural system such as the presence of a fractal dimension combined with a "rich" spectral signature.

Natural systems operate close to their functional capacity. Combining the ideas of computing networks and pseudo-physical systems creates a new paradigm. This new paradigm captures the behavior of computing networks for the purpose of improved communication and processing capability.

Extensive research, comparing distributed computing systems to natural ecologies, alludes to the link between order in systems and "chaos." In many natural systems, chaos implies order.

The major contribution of this research is the "discovery" of chaos in computing networks by means of an innovative technique: the combination of the Correlation Dimension and the Fast Fourier Transform. The implication of the research is it opens a new area of study which should lead to "physically-friendly" computing networks. The conclusion of this research is that the measurement of chaos in computing networks implies order (yet unused).

Share

COinS