In International Conference on Distributed Systems (ICDCS), June 2012. IEEE.
Acceptance rate: 13%.
Publish/Subscribe (P/S) systems and file sharing applications traditionally share the common goal of disseminating data among large populations of users. Despite this similarity, the former focuses on timely dissemination of small-sized events or notification messages, while the latter presumes much larger types of bulk content with less emphasis on the time needed between release and delivery of data. In this paper, we develop a push-based content dissemination mechanism to bridge this gap by adopting the P/S model for timely distribution of bulk data amongst clients based on their subscription interests. We develop a hybrid two-layer architecture in which P/S brokers act as coordinators and guide their clients with similar interests to engage in direct exchange of data blocks in a peer-to-peer and cooperative fashion. Furthermore, we use network coding in order to facilitate exchange of data among clients. Our peer-assisted approach offloads the burden of disseminating huge volumes of data from P/S brokers to subscribers themselves. As a proof of concept, we have implemented our approach and carried out large-scale experimental evaluation on a cluster with up to 1000 subscribers. Our results indicate that with a capped client upload bandwidth of 200 KB/s, 1 TB worth of aggregate data can be disseminated among 1000 subscribers in under 2 hours. Finally, as an added advantage of our approach, brokers deploy strategies that can help shape traffic flows in multi-domain network settings.