Acknowledgements

Support for the evaluation was provided by SyncFree and PlanetLab. I would like to thank Marek Zawirsky for his support to integrate partial replication into SwiftCloud. Thanks to Manuel Bravo for his useful remarks and help throughout the writing of this master’s thesis. Special thanks to Xavier Bellekens who read versions of the text and gave sage advice. Thanks to reddit inc. for their website, releasing their source code, and providing API access to their data. I am grateful for my readers, Professor Marc Lobelle and Zhongmiao Li, for their time. Finally I would like to thank my supervisor, Professor Peter Van Roy, for his comments and engagement. Current mobile and web applications replicate data increasingly at the client-side to reduce latency and to tolerate network issues. Mobile devices have however limited storage and network capabilities. Thus large data structures at the server-side need to be split before being sent to the client and reintegrated when modified, in order to keep a low memory and network usage. Ad-hoc solutions tend to be poorly integrated with server-side storage, and do not have well-defined consistency guarantees. We propose a replication mechanism built upon conflict-free replicated data types (CRDT) to seamlessly replicate parts of large data structures. We define partial replication and give an approach to keep the strong eventual consistency properties of CRDTs with par-tial replicas. We integrate our mechanism into SwiftCloud, a transactional system that brings geo-replication to the client. We evaluate the solution with a content-sharing ap-plication. Our results show improvements in bandwidth, memory, and latency usage over both classical geo-replication and the existing SwiftCloud solution.