Date of Award

Spring 2004

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

Symeon Papavassiliou

Fourth Advisor

Teunis J. Ott

Fifth Advisor

Lev A. Zakrevski


With the development of the Internet, Internet service providers (ISPs) are required to offer revenue-generating and value-added services instead of only providing bandwidth and access services. Virtual Private Network (VPN) is one of the most important value-added services for ISPs.

The "classical" VPN service is provided by implementing layer 2 technologies, either Frame Relay (FR) or Asynchronous Transfer Mode (ATM). With FR or ATM, virtual circuits are created before data delivery. Since the bandwidth and buffers are reserved, the QoS requirements can be naturally guaranteed. In the past few years, layer 3 VPN technologies are widely deployed due to the desirable performance in terms of flexibility, scalability and simplicity. Layer 3 VPNs are built upon IP tunnels, e.g., by using PPTP, L2TP or IPSec. Since IP is "best-of-effort" in nature, the QoS requirement cannot be guaranteed in layer 3 VPNs. Actually, layer 3 VPN service can only provide secure connectivity, i.e., protecting and authenticating IP packets between gateways or hosts in a VPN. Without doubt, with more applications on voice, audio and video being used in the Internet, the provision of QoS is one of the most important parts of the emerging services provided by ISPs. An intriguing question is: "Is it possible to obtain the best of both layer 2 and 3 VPN? Is it possible to provide guaranteed or predictable QoS, as in layer 2 VPNs, while maintaining the flexibility and simplicity in layer 3 VPN?" This question is the starting point of this study.

The recently proposed hose model for VPN possesses desirable properties in terms of flexibility, scalability and multiplexing gain. However, the "classic" fair bandwidth allocation schemes and weighted fair queuing schemes raise the issue of low overall utilization in this model. A new fluid model for provider-provisioned virtual private network (PPVPN) is proposed in this dissertation. Based on the proposed model, an idealized fluid bandwidth allocation scheme is developed. This scheme is proven, analytically, to have the following properties: 1) maximize the overall throughput of the VPN without compromising fairness; 2) provide a mechanism that enables the VPN customers to allocate the bandwidth according to their requirements by assigning different weights to different hose flows, and thus obtain the predictable QoS performance; and 3) improve the overall throughput of the ISPs' network. To approximate the idealized fluid scheme in the real world, the 2-dimensional deficit round robin (2-D DRR and 2-D DRR+) schemes are proposed. The integration of the proposed schemes with the best-effort traffic within the framework of virtual-router-based VPN is also investigated. The 2-D DRR and 2-D DER-+ schemes can be extended to multi-dimensional schemes to be employed in those applications which require a hierarchical scheduling architecture. To enhance the scalability, a more scalable non-per-flow-based scheme for output queued switches is developed as well, and the integration of this scheme within the framework of the MPLS VPN and applications for multicasting traffics is discussed. The performance and properties of these schemes are analyzed.