Exploring data replication

Year 2018
Author Pierre-Louis Roman
Publisher Universite De Rennes
Link View Research Paper


If you have ever lost your personal data due to a machine failure, you know how important backups and data replication are. For service suppliers, backups are vital for the robustness of their services, as losing their user’s data is not an option. Data is therefore usually replicated in several geo distributed regions to avoid even the worst failures.

However, in this situation, each update must also be replicated by every instance to maintain consistent replicas, which rapidly gets more difficult as the number of users, and hence updates, increases. The spectacular growth of the number of people connected to the internet and the increasing time they spend on it clearly lay a challenge to data replication. This challenge and its ramifications lay at the heart of this paper.

Download it now to discover a novel eventual consistency mechanism that offers heterogeneous properties in terms of data consistency and delivery latency. Primary nodes can deliver updates faster at the cost of a small consistency penalty, while Secondary nodes experience stronger consistency but with a slightly higher latency. Both sets of nodes observe a consistent state with high probability once dissemination completes. Discover what it all means for data replication in the world of decentralisation!