Matchmaker: signaling for dynamic publish/subscribe applications

Abstract

The publish/subscribe (pub/sub) paradigm provides content-oriented data dissemination in which communication channels are established between content publishers and content subscribers based on a matching of subscribers interest in the published content provided - a process we refer to as "matchmaking". Once an interest match has been made, content forwarding state can be installed at intermediate nodes (e.g., active routers, application-level relay nodes) on the path between a content provider and an interested subscriber. In dynamic pub/sub applications, where published content and subscriber interest change frequently the signaling overhead needed to perform matchmaking can be a significant overhead. We first formalize the matchmaking process as an optimization problem, with the goal of minimizing the amount of matchmaking signaling messages. We consider this problem for both shared and per-source multicast data (content) distribution topologies. We characterize the fundamental complexity of the problem, and then describe several efficient solution approaches. The insights gained through our analysis are then embodied in a novel active matchmaker signaling protocol (AMSP). AMSP dynamically adapts to applications' changing publication and subscription requests through a link-marking approach. We simulate AMSP and two existing broadcast-based approaches for conducting matchmaking, and find that AMSP significantly reduces signaling overhead.