UC San Diego

As Internet traffic continues to grow unabated at an exponential rate, it is unclear whether or not the existing packet routing network architecture based on electronic routers will continue to scale at the necessary pace. On the other hand, optical fiber and switching elements have demonstrated an abundance of capacity that appears to be unmatched by electronic routers. In particular, the simplicity of circuit switching makes it well-suited for optical implementations. Therefore, given the rapidly increasing traffic and optical transport capabilities, and the growing disparity between optical capabilities and Moore's Law, we would like to bridge this gap for future optical networks. In this thesis, we present a new approach to optical networking based on a paradigm of "coarse optical circuit switching by default" and "adaptive rerouting over circuits with spare capacity". We consider the provisioning of long-duration quasi-static optical circuits between edge routers at the boundary of the network to carry the traffic by default. When the provisioned circuit is inadequate, excess traffic demand is rerouted through circuits with spare capacity. In particular, by adaptively load-balancing across circuits with spare capacity, excess traffic is routed on their final destinations without the need to create circuits "on-the-fly". We call this new network architecture COPLAR, which stands for "[C]oarse [OP]tica[L] circuit switching with [A]daptive [R]erouting". Specifically, we first formulate the problem of circuit provisioning as a multipath utility max-min bandwidth allocation problem, which considers routing as an optimization parameter rather than input. The optimal and local algorithms proposed in the thesis are the first solutions to this problem. For traffic that cannot be handled by the default provisioned circuits, we apply an adaptive routing algorithm based on game theory to adaptively reroute the excess traffic over circuits with spare capacities. To evaluate COPLAR, we conducted extensive experiments using two separate real large-ISP PoP (point of presence)-level topologies, Abilene and GEANT. The results show that coplar has the ability to improve network throughput while significantly reducing electronic router overhead and the number of O/E/O conversions in the network. Finally, we show that our utility max-min bandwidth allocation algorithm can be extended to other network problems. In particular, we present a network security application called PSP (Proactive Surge Protection) that provides a defense against bandwidth-based distributed denial-of-service attacks