Recently one has witnessed the emergence of several service-based overlay architectures aimed at providing service qualities better than the best effort service available in today's Internet. Each of these architectures addresses different aspects of service quality issues, including traffic engineering, Quality of Service (QoS), and fault tolerance. However, due to the fact that resource availability and topology of the underlying network layer are hidden from the overlay, these architectures have to use various techniques to gain better control of the resources that may be used by the overlay, making it difficult to integrate different architectures to provide rich service qualities for various applications.
In this proposal, a novel approach is presented which provides a unified framework to enable traffic engineering, QoS, and fault tolerance. The PIs will develop decentralized traffic control laws that will provide all these service qualities simultaneously. Since these control laws only use source inferable information, the added features are implemented locally at the source nodes. Moreover, since these control laws rely on results from the theory of optimization and control, they achieve multiple design objectives with provable global stability and convergence properties.
Since the approach taken complies with the end-to-end design principle, it allows: Scalable and fast development of applications requiring various service quality features; Quick adaptation to the existing overlay and IP networking architectures to enhance their service quality feature sets; Better understanding, from a more fundamental point of view, of the existing end-to-end control protocols, such as TCP with or without multiple paths.