Large-scale reliable multicast of small messages

Several applications could benefit from large-scale reliable multicast of small messages. For example, the problems of propagating invalidations for web-cache consistency, dissemination of stock quotes to traders, and propagation of information about web updates to search engines all require reliable multicast of small messages. While multicasting is itself an active area of research, and each of the problems mentioned have received attention in literature, there are no guiding principles for facilitating multicast of small messages. First, we outline a global rendezvous architecture (GRA) - an application-level architecture for large-scale reliable multicast of small messages. The main contributions of GRA are the global rendezvous point concept, the join protocol for client, and the facility for trust across AS boundaries. Second, we describe QuickFlow, an architecture for web cache consistency that utilizes GRA, and show the benefits of QuickFlow compared to previous approaches. Third, we show how FreshFlow, an architecture proposed for search engine freshness, follows GRA. We also present new results for FreshFlow that demonstrate the scalability of FreshFlow.