Date of Award

Fall 2003

Document Type


Degree Name

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


Electrical and Computer Engineering

First Advisor

Nirwan Ansari

Second Advisor

Jianguo Chen

Third Advisor

Edwin Hou

Fourth Advisor

Sirin Tekinay

Fifth Advisor

Lev A. Zakrevski


IP-based virtual private networks (VPNs) have the potential of delivering cost-effective, secure, and private network-like services. Having surveyed current enabling techniques, an overall picture of IP VPN implementations is presented.

In order to provision the equivalent quality of service (QoS) of legacy connection-oriented layer 2 VPNs (e.g., Frame Relay and ATM), IP VPNs have to overcome the intrinsically best effort characteristics of the Internet. Subsequently, a hierarchical QoS guarantee framework for IP VPNs is proposed, stitching together development progresses from recent research and engineering work.

To differentiate IP VPN QoS, the proportional QoS differentiation model, whose QoS specification granularity compromises that of IntServ and Diffserv, emerges as a potential solution. The investigation of its claimed capability of providing the predictable and controllable QoS differentiation is then conducted.

With respect to the loss rate differentiation, the "packet shortage" phenomenon shown in two classical proportional loss rate (PLR) dropping schemes is studied. On the pursuit of a feasible solution, the potential of compromising the system resource, that is, the buffer, is ruled out; instead, an enhanced "debt-aware" mechanism is suggested to relieve the negative effects of "packet shortage." Simulation results show that "debt-aware" partially curbs the biased loss rate ratios, and improves the queueing delay performance as well.

With respect to the delay differentiation, the dynamic behavior of the average delay difference between successive classes is first analyzed, aiming to gain insights of system dynamics. Then, two classical delay differentiation mechanisms, that is,proportional average delay (PAD) and waiting time priority (WTP), are simulated and discussed. Based on observations on their differentiation performances over both short and long time periods, a combined delay differentiation (CDD) scheme is introduced. Simulations are utilized to validate this method.

Both loss and delay differentiations are based on a series of differentiation parameters. Though previous work on the selection of delay differentiation parameters has been presented, that of loss differentiation parameters mostly relied on network operators' experience. A quantitative guideline, based on the principles of queueing and optimization, is then proposed to compute loss differentiation parameters. Aside from analysis, the new approach is substantiated by numerical results.