Ph. D. Thesis

Flow-Aware Networking (FAN) architecture is evaluated in this dissertation with regard to its service differentiation and quality of service (QoS) assurance capabilities. The network neutrality debate is presented first, and it is shown that the potential resolutions will have a strong impact on QoS architectures. It is also shown that FAN, as well as all the proposed mechanisms, is perfectly suited to the future Internet and in-line with the network neutrality principle. Secondly, the detailed concept of FAN is presented and compared with other flow-based QoS architectures designed for IP networks. It is argued that all the solutions have their advantages and disadvantages, however, in FAN, the pros outnumber the cons in comparison with other architectures. Not only is it net neutrality compliant, but also efficient and scalable.

The main goal of the dissertation is to propose and evaluate new mechanisms to enhance service differentiation capabilities of FAN architecture. The waiting times phenomenon as a result of admission control functionality is documented. Next, differentiated blocking and differentiated queuing mechanisms are proposed. Those mechanisms offer improved prospects of providing differentiated treatment for end-user flows. The Static Router Configuration approach is also presented, as a feasible method of implementing the new mechanisms. Finally, Class of Service on Demand is shown as the ultimate method of providing rich service differentiation without violating the network neutrality principle.

Although FAN offers QoS protection, the basic method is inefficient. This leads to fair rate degradations shown and analyzed in-depth. Several solutions to the problem are presented in the dissertation. The static limitation mechanism is proposed as a simple, yet efficient way of improving service assurance. It is shown that the limitation mechanism significantly contributes to FAN's scalability and yields great performance benefits. The static mechanism can be enhanced by the dynamic limitation mechanism which offers better results, although, only provided that the mechanism is properly set up, which is not a trivial task. To overcome this drawback, an automatic intelligent limitation mechanism is proposed which can adjust to current network conditions and is not dependent on the proper setup.

Finally, the predictive approach is presented, which changes the functioning of the admission control block in FAN. Instead of waiting for congestion to appear and only then blocking new connections, the mechanism takes a pro-active approach and starts to act on the basis of the predicted values of the congestion indicators. This enables the admission control block to react appropriately even before congestion occurs. It is shown that the best results are obtained when the predictive approach is combined with the limitation mechanism.