The Use of IP - Anycast to Construct Efficient Multicast Trees by Dina Katabi

This thesis consists of three parts: the design of a framework for an Internet-wide IP anycast service, the application of this framework in optimizing the cost of core based multicast trees, and the development of a new interdomain multicast protocol. The anycast framework solves the intrinsic scalability problem associated with a global IP anycast service. It does so by dividing the interdomain anycast routing system into two modules. The first module provides low-cost anycast routes. The second module, run by the end domains, generates enhanced anycast routes that are customized according to the beneficiary domains interests. The design uses an adaptive caching system plus a route learning mechanism. The second part of the thesis uses the IP anycast service to extend the Core-BasedTree multicast protocol (CBT) with the option to build optimized cost dynamic-core trees. These trees are free from the two major drawbacks shown by CBTs poor core placement and traffic concentration around the core router. The proposed extension makes all on-tree routers join the same anycast group such that the multicast tree appears to all off-tree routers as one network entity, which can be addressed directly. We use this property to direct data packets and join requests along the shortest path to the tree decreasing bandwidth consumption and join latency. Since the resulting tree is simultaneously an anycast group and a core based tree, we call it an ACBT. Our simulation shows that ACBT reduces bandwidth consumption by a factor of 25 In the last part of this thesis, we develop FleSc a flexible and scalable interdomain multicast protocol. Our protocol builds on the BGMP model to establish bi-directional trees rooted at an owner domain. However, our design differs from BGMP in two important ways. First, it uses ideas from our anycast framework to favor short local branches over hierarchical long ones. Second, the established tree uses hard state, which prevents transient loops and facilitates the establishment of alternative branches when the default ones do not accommodate the desired QoS. The architecture is scalable since it uses a small set of hierarchical aggregatable multicast routes that are extended at each end domain with a set of on-demand generated routes, which are customized according to the domains interests. It is flexible since it provides a variety of possible branches to chose from the one that reduces band-