Thesis (M.S. in Operations Research)--Naval Postgraduate School, March 2010. / Thesis Advisor(s): Alderson, David L. Second Reader: Carlyle, W. Matthew. "March 2010." Description based on title screen as viewed on April 23, 2010. Author(s) subject terms: Internet Service Provider, traffic engineering, network provisioning, network design, network topology, demand matrices Includes bibliographical references (p. 87-89). Also available in print.

The increasing demand for real-time multimedia applications and the lack of quality of service (QoS) support in public best-effort or Internet Protocol (IP) networks has prompted many researchers to propose improvements on the QoS of such networks. This research aims to improve the QoS of real-time multimedia applications in public best-effort networks, without modifying the core network infrastructure or the existing codecs of the original media applications. A source buffering control is studied based on a fluid model developed for a single flow transported over a best-effort network while allowing for flow reversal. It is shown that this control is effective for QoS improvement only when there is sufficient flow reversal or packet reordering in the network. An alternate control strategy based on predictive multi-path switching is studied where only two paths are considered as alternate options. Initially, an emulation study is performed, exploring the impact of path loss rate and traffic delay signal frequency content on the proposed control. The study reveals that this control strategy provides the best QoS improvement when the average comprehensive loss rates of the two paths involved are between 5% and 15%, and when the delay signal frequency content is around 0.5 Hz. Linear and nonlinear predictors are developed using actual network data for use in predictive multi-path switching control. The control results show that predictive path switching is better than no path switching, yet no one predictor developed is best for all cases studied. A voting based control strategy is proposed to overcome this problem. The results show that the voting based control strategy results in better performance for all cases studied. An actual voice quality test is performed, proving that predictive path switching is better than no path switching. Despite the improvements obtained, predictive path switching control has some scalability problems and other shortcomings that require further investigation. If there are more paths available to choose from, the increasing overhead in probing traffic might become unacceptable. Further, if most of the VoIP flows on the Internet use this control strategy, then the conclusions of this research might be different, requiring modifications to the proposed approach. Further studies on these problems are needed.