A hybrid publish subscribe protocol

Abstract

Content-based publish/subscribe system performance depends upon the efficient subscription matching and event dissemination to interested subscribers. We propose a hybrid content-based publish/subscribe protocol for large size events wherein a centralized brokering system is coupled with a decentralized BitTorrent-like peer-to-peer (P2P) protocol for scalable event distribution among publishers and subscribers. Events are mapped to a torrent that grows as new events are published. Subscribers self-broker on event metadata and request content only if interested. Subscriber interests determine event popularity that the broker estimates with sampling. Popular events are disseminated P2P; unpopular events, directly from the broker; and somewhat popular ones, with P2P and broker-directed pre-seeding. The challenge is the dissemination of popular events without overwhelming centralized resources while efficiently disseminating unpopular events that lack sufficient interest to sustain gossip-based dissemination. The key advances include new means of handling variable event popularity inherent in content-based pub/sub and an adaptive anti-entropy mechanism for undelivered events.