Quality-of-Service Routing for Voice-over-IP in Service Overlay Networks

Voice-over-IP (VoIP) becomes more and more popular with the development of service convergence in the next generation network. This thesis focuses on improving VoIP quality with application-layer routing in service overlay networks.;Based on the investigation of the Internet end-to-end delay, we find that VoIP quality on a pair of diverse paths is better and more stable than that on a single path. We therefore propose a novel centralized data fusion approach to search for the best pair of diverse paths. This method jointly optimizes source routing with adaptive play-out scheduling at the receiver. It requires transmitting the delay distributions of all the overlay links for estimating the delay distributions of diverse paths. We propose to transmit only the model parameters of the link delay distributions to reduce the communication overhead. It is shown that the best pair of diverse paths can be estimated with a small error.;Nonetheless, the centralized approach is computationally expensive. We therefore propose an online diverse routing method, which uses distributed learning automata to actively probe path delays and to determine the best pair of diverse paths for VoIP based on the state of the learning automata. We have demonstrated the scalability and the optimality of the approach by simulations, and proven the optimality of the approach using Kushner's weak convergence method. VoIP quality has been shown to improve from unsatisfactory levels to satisfactory levels. In addition, we propose a method to detect and recover from link failures based on the state of the learning automata. Considerable improvement in link failure recovery time has been achieved.;In sum, this work demonstrates that the proposed centralized diverse routing approach is effective to improve VoIP quality in terms of R-factor for small overlay networks, and that the proposed distributive diverse routing approach together with the link failure detection scheme provides a scalable, effective and robust solution to VoIP routing for large overlay networks.;Internet end-to-end delay is one of the most important impairments on VoIP quality. We therefore analyze, model and simulate it to better understand it and thus to discover potential advantages of routing in service overlay networks.