Date of Award

Spring 2004

Document Type

Dissertation

Degree Name

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

Department

Electrical and Computer Engineering

First Advisor

Symeon Papavassiliou

Second Advisor

Nirwan Ansari

Third Advisor

Roberto Rojas-Cessa

Fourth Advisor

Mostafa Hashem Sherif

Fifth Advisor

MengChu Zhou

Abstract

Next Generation IP technology has the potential to prevail, both in the access and in the core networks, as we are moving towards a multi-service, multimedia and high-speed networking environment. Many new applications, including the multimedia applications, have been developed and deployed, and demand Quality of Service (QoS) support from the Internet, in addition to the current best effort service. Therefore, QoS provisioning techniques in the Internet to guarantee some specific QoS parameters are more a requirement than a desire. Due to the large amount of data flows and bandwidth demand, as well as the various QoS requirements, scalability and fine granularity in QoS provisioning are required. In this dissertation, the end-to-end QoS provisioning mechanisms are mainly studied, in order to provide scalable services with fine granularity to the users, so that both users and network service providers can achieve more benefits from the QoS provisioned in the network.

To provide the end-to-end QoS guarantee, single-node QoS provisioning schemes have to be deployed at each router, and therefore, in this dissertation, such schemes are studied prior to the study of the end-to-end QoS provisioning mechanisms. Specifically, the effective sharing of the output bandwidth among the large amount of data flows is studied, so that fairness in the bandwidth allocation among the flows can be achieved in a scalable fashion. A dual-rate grouping architecture is proposed in this dissertation, in which the granularity in rate allocation can be enhanced, while the scalability of the one-rate grouping architecture is still maintained. It is demonstrated that the dual-rate grouping architecture approximates the ideal per-flow based PFQ architecture better than the one-rate grouping architecture, and provides better immunity capability.

On the end-to-end QoS provisioning, a new Endpoint Admission Control scheme for Diffserv networks, referred to as Explicit Endpoint Admission Control (EEAC), is proposed, in which the admission control decision is made by the end hosts based on the end-to-end performance of the network. A novel concept, namely the service vector, is introduced, by which an end host can choose different services at different routers along its data path. Thus, the proposed service provisioning paradigm decouples the end-to-end QoS provisioning from the service provisioning at each router, and the end-to-end QoS granularity in the Diffserv networks can be enhanced, while the implementation complexity of the Diffserv model is maintained. Furthermore, several aspects of the implementation of the EEAC and service vector paradigm, referred to as EEAC-SV, in the Diffserv architecture are also investigated. The performance analysis and simulation results demonstrate that the proposed EEAC-SV scheme, not only increases the benefit to the service users, but also enhances the benefit to the network service provider in terms of network resource utilization. The study also indicates that the proposed EEAC-SV scheme can provide a compatible and friendly networking environment to the conventional TCP flows, and the scheme can be deployed in the current Internet in an incremental and gradual fashion.

Share

COinS