Internet traffic engineering

Due to the dramatically increasing popularity of the services provided over the public Internet, problems with current mechanisms for control and management of the Internet are becoming apparent. In particular, it is increasingly clear that the Internet and other networks built on the Internet protocol suite do not provide sufficient support for the efficient control and management of traffic, i.e. for Traffic Engineering. This dissertation addresses the problem of traffic engineering in the Internet. It argues that traffic management techniques should be applied at multiple timescales, and not just at data timescales as is currently the case. It presents and evaluates mechanisms for traffic engineering in the Internet at two further timescales: flow admission control and control of per-flow packet marking, enabling control timescale traffic engineering; and support for load based inter-domain routeing in the Internet, enabling management timescale traffic engineering. This dissertation also discusses suitable policies for the application of the proposed mechanisms. It argues that the proposed mechanisms are able to support a wide range of policies useful to both users and operators. Finally, in a network of the size of the Internet consideration must also be given to the deployment of proposed solutions. Consequently, arguments for and against the deployment of these mechanisms are presented and the conclusion drawn that there are a number of feasible paths toward deployment. The work presented argues the following: firstly, it is possible to implement mechanisms within the Internet framework that enable traffic engineering to be carried out by operators; secondly, that applying these mechanisms with suitable policies can ease the management problems faced by operators and at the same time improve the efficiency with which the network can be run; thirdly, that these improvements can correspond to increased network performance as viewed by the user; and finally, that not only the resulting deployment but also the deployment process itself are feasible.