New control-plane architecture for QoS-guaranteed Internet

The Internet is evolving to become the ubiquitous network and needs to provide the quality-of-service (QoS) levels that the end-users are accustomed to. QoS requirements can be achieved with components implemented in different layers and various domains - control and data planes. Effective implementations of data-plane components are well understood and available. They only need local information in a router or switch. In contrast, control-plane components, such as FAC (flow admission control) and network provisioning, need the state information of the entire network and they are the main challenges in the design of the future QoS Internet. The main objective of our work is to explore new ideas and solutions for several key control-plane components, including flow admission control, routing, and network dimensioning. The foundation of the proposed solutions is the concept of 'nonblocking backbone network'. A network is called nonblocking if it can always admit a flow without causing congestion on any link inside the network as long as adding the flow does not violate the given traffic constraint. The major feature of the nonblocking backbone network is that the information required to make FAC decisions can be entirely localized, and a distributed and effective FAC scheme can be easily designed. By integrating the proposed schemes into existing routers' control-plane architecture, we intend to show that hard QoS can be supported in a large scale and cross domain boundaries in the Internet. The key issue in designing a nonblocking backbone network is to determine how much traffic can be admitted to a network without violating the QoS requirement of the traffic. We develop routing and load-distribution schemes for MPLS-based and IP-based backbone networks. Both optimal and efficient heuristic route computation algorithms have been proposed for these types of networks. We also study the issue of designing a restorable MPLS-based VPN network with QoS guarantee. The developed techniques can also be applied to networks where finding optimal routing for a class of traffic matrices, instead of for a single traffic matrix, is required