On predictable large-scale data delivery in prefix-based virtualized content networks

Abstract

IPTV, software replication, and other large scale content distribution services raise the need for fast and efficient content delivery mechanisms in underlay as well as overlay networks. Multicast, the natural approach on the network layer, has not been deployed globally, and solutions are pushed to the application layer. For a flexible, sustainable deployment the distribution mechanisms in use should scale up to many thousand group members and provide predictable performance to dynamically adjust to actual performance requirements. In this paper, we present a rigorous analytical model complemented by extensive simulations for content delivery on prefix-based overlay trees. We examine BIDIR-SAM, a generic multicast distribution scheme guided by prefixes that allow for late next-hop binding. Our evaluation quantitatively substantiates all major performance aspects. We derive the distribution functions of hop counts, packet replication loads, as well as all relevant cost measures, which scale logarithmically with network and receiver sizes. Prefix-based content delivery exhibits a churn resistance similar to the underlying key-based routing layer and a multicast efficiency scaling factor close to native group communication protocols. These results make the approach especially suitable for large and very large content groups.