Date of Award

Fall 2001

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Electrical Engineering - (Ph.D.)

Department

Electrical and Computer Engineering

First Advisor

Sirin Tekinay

Second Advisor

Ali N. Akansu

Third Advisor

Constantine N. Manikopoulos

Fourth Advisor

Jay Jorgenson

Fifth Advisor

Nirwan Ansari

Abstract

The provision of QoS to applications traffic depends heavily on how different traffic types are categorized and classified, and how the prioritization of these applications are managed. Bandwidth is the most scarce network resource. Therefore, there is a need for a method or system that distributes an available bandwidth in a network among different applications in such a way that each class or type of traffic receives their constraint QoS requirements.

In this dissertation, a new renegotiation based dynamic resource allocation method for variable bit rate (VBR) traffic is presented. First, pros and cons of available off-line methods that are used to estimate selfsimilarity level (represented by Hurst parameter) of a VBR traffic trace are empirically investigated, and criteria to select measurement parameters for online resource management are developed. It is shown that wavelet analysis based methods are the strongest tools in estimation of Hurst parameter with their low computational complexities, compared to the variance-time method and R/S pox plot. Therefore, a temporal energy distribution of a traffic data arrival counting process among different frequency sub-bands is considered as a traffic descriptor, and then a robust traffic rate predictor is developed by using the Haar wavelet analysis. The empirical results show that the new on-line dynamic bandwidth allocation scheme for VBR traffic is superior to traditional dynamic bandwidth allocation methods that are based on adaptive algorithms such as Least Mean Square, Recursive Least Square, and Mean Square Error etc. in terms of high utilization and low queuing delay. Also a method is developed to minimize the number of bandwidth renegotiations to decrease signaling costs on traffic schedulers (e.g. WFQ) and networks (e.g. ATM). It is also quantified that the introduced renegotiation based bandwidth management scheme decreases heavytailedness of queue size distributions, which is an inherent impact of traffic self similarity.

The new design increases the achieved utilization levels in the literature, provisions given queue size constraints and minimizes the number of renegotiations simultaneously. This renegotiation -based design is online and practically embeddable into QoS management blocks, edge routers and Digital Subscriber Lines Access Multiplexers (DSLAM) and rate adaptive DSL modems.

Share

COinS