Feverfew: a scalable coverage-based hybrid overlay for Internet-scale pub/sub networks

The publish/subscribe (pub/sub) paradigm is a popular communication model for data dissemination in large-scale distributed networks. However, scalability comes with a contradiction between the delivery latency and the memory cost. On one hand, constructing a separate overly per topic guarantees real-time dissemination, while the number of node degrees rapidly increases with the number of subscriptions. On the other hand, maintaining a bounded number of connections per node guarantees small memory cost, while each message has to traverse a large number of uninterested nodes before reaching the subscribers. In this paper, we propose Feverfew, a coverage-based hybrid overlay that disseminates messages to all subscribers without uninterested nodes involved in, and increases the average number of node connections slowly with an increase in the number of subscribers and nodes. The major novelty of Feverfew lies in its heuristic coverage mechanism implemented by combining a gossip-based sampling protocol with a probabilistic searching protocol. Based on the practical workload, our experimental results show that Feverfew significantly outperforms existing coverage-based overlay and DHT-based overlay in various dynamic network environments.