LIFT: Logical Information Feedback Tree for Information Dissemination in Wide-Area Networks

A key challenge to the deployment of a number of interesting networking technologies, such as content distribution networks, application layer multicast and reliable multicast protocols, is the efficient many-to-many distribution of data using network overlays. The use of network overlays has become increasingly popular due to the potential problems associated with native IP multicast and the desire to modify, route, or process data as it traverses the network. In this paper, we introduce the Logical Information Feedback Tree (LIFT) for dynamically building and maintaining completely decentralized, self-organizing, balanced network overlays across wide-area nodes for the purpose of information dissemination. The goal of LIFT is to provide a network overlay that: i) is highly optimized in terms of some global network metric such as perceived network latency; ii) converges quickly to a maximum width bounded by O(lg n); iii) exhibits low stress on the underlying substrate network; iv) is built without any centralized control or information so that it can scale to large numbers of nodes spread over a wide area; and v) is robust to dynamic insertion and deletion of nodes. We evaluate LIFT against these metrics through simulation in ns, a popular network simulator. Our results indicate that LIFT provides an efficient mechanism for dynamically creating network overlays that can support scalable information dissemination for a variety of applications.