Internet traffic engineering: QoS translation and survivability

The problem of traffic engineering has been widely recognized as critical to the development of operational Internet. Traffic engineering involves evaluation and optimization of the performance of operational IP networks. This thesis studies the performance evaluation problem from the perspectives of the users and the providers. Scalability requirements necessitate the use of an analytical framework to translate network level service quality in terms of the performance of user level applications. This thesis presents an important building block for such a framework in the form of an analytical model for TCP throughput prediction using easily available input information. Evaluating the performance of the networks from the providers' perspective naturally involves taking into account the service quality requirements of the users. The ever-increasing dependence on the Internet has made it important that the Internet service can survive the failures in the networks. The cost of survivability and the speed of recovery constitute the two important aspects of the network survivability. This thesis evaluates the cost of network operation for the providers taking into account the survivability requirements. This is followed by an investigation into the role of simple topological modifications in reducing the cost of survivability. The speed of recovery problem is studied from the perspective of different time scales. At the macro time scale, the objective is to enable fast detection and correction of network failures by the network administrators. This thesis compares the two most relevant alternatives for the purpose. At the micro time scale, the objective is fast rerouting of the traffic around the failures by the in-built survivability mechanisms while the failures have not yet been corrected. This thesis investigates the operation of popular survivability mechanisms operating at the IP and optical layers and suggests ways and means of improving their speed of recovery from network failures.