Large-scale decentralized topic-based publish/subscribe systems are increasingly becoming important. Many online applications could benefit from such a system. An overlay to connect peers and disseminate messages based on their interest is fundamental to several online applications including but not limited to decentralized online social networks, media streaming applications like IPTV, Spotify, News aggregation systems like RSS feeds, software package update systems for operating systems etc.
One of the main challenges dealing with decentralized systems is providing a reliable service. In P2P applications, since the network is dynamic it is important to design strategic choice of neighbors to ensure high availability of the service. To this end we have ongoing work which proposes a gossip-based overlay construction for decentralized topic-based pub/sub. In this approach each node in the system chooses its neighbors using random peer sampling service[1], based on the topics the node is interested in. However, the choice of neighbors purely based on topics of interest does not guarantee high availability.
The motivation of this work is to devise a strategy for choice of neighbors in such a way that, with high probability, at any time, there are always enough nodes available as neighbors to keep the overlay connected. Several possible approaches to solving this problem can be explored during the course of the thesis.
[1] S. Voulgaris, D. Gavidia, and M. van Steen, “CYCLON: Inexpensive Membership Management for Unstructured P2P Overlays,” Network and Systems Management, June 2005.