Defense committee: Prof. Badri Nath (chair), Prof. Richard Martin, Prof. Thu Nguyen and Dr. Arup Acharya (IBM T.J. Watson Research Center)

Abstract:

The scalability limitations of the stateful QoS architectures have led to the popularity of stateless architectures where per-flow states are maintained only at the edge routers. Like any QoS architecture, core-stateless architectures also require a corresponding admission control framework to ensure that a reservation request is granted only if the network has sufficient resources to support it. Admission control is a difficult problem in core-stateless QoS architectures, especially if we want to support guaranteed services.

The existing admission control methods to support core-stateless services that require bandwidth guarantees, rely on a centralized broker node. This dissertation proposes and evaluates a novel admission control framework for such stateless QoS architectures. Our architecture is fully-distributed, guarantees that a request is admitted only if there is sufficient bandwidth to support it, and maintains a high network utilization in high bandwidth environments. In our architecture, the ingress edge router of a request takes admission decision for all links on its edge-to-edge path. To achieve a high utilization, we first establish the ideal state that any edge-router based distributed admission control method should attain. Then we design a novel protocol which tries to attain the ideal state while adapting to the dynamic traffic conditions. We also show how our resource management framework could be used with novel signaling techniques to design a light-weight admission control architecture for heterogeneous multicast.

Another contribution of this dissertation is in defining the notion of e-fairness to reduce the overhead incurred attaining max-min fairness. The ideal resource management state depends on fair bandwidth partitioning. This allows us to use e-fairness to reduce the overhead incurred in the resource management portion of our admission control framework.