Feres: Flexible and Efficient Replica Synchronization For Diverse Environments

This paper presents Feres, a peer-to-peer data synchronization protocol that can be used to construct new, flexible distributed file systems that share data across collections of devices with limited, varying, or intermittent connectivity. The amount of flexibility provided by Feres is not matched by existing protocols. In particular, Feres allows any device to synchronize any subset of data with any other node, provides options to chose either data or meta-data and log-based or state-based synchronization, detects conflicts and supports application specific commit policies without require rollback. In addition, the overheads associated with Feres are reasonable. Because Feres is flexible, it can send the right data via the right network paths and dramatically outperform less flexible traditional client-server or server-replication protocols – we observe up to 15 times speedup for one scenario. At the same time, Feres is more efficient than existing similarly flexible protocols like PRACTI – we observe order of magnitude of bandwidth savings in some experiments.