State-of-the-art fault-tolerant and federated data management systems rely on fully-replicated designs in which all participants have equivalent roles. Consequently, these systems have only limited scalability and are ill-suited for high-performance data management. As an alternative, we propose a hierarchical design in which a Byzantine cluster manages data, while an arbitrary number of learners can reliable learn these updates and use the corresponding data. To realize our design, we propose the delayed-replication algorithm, an efficient solution to the Byzantine learner problem that is central to our design. The delayed-replication algorithm is coordination-free, scalable, and has minimal communication cost for all participants involved. In doing so, the delayed-broadcast algorithm opens the door to new high-performance fault-tolerant and federated data management systems. To illustrate this, we show that the delayed-replication algorithm is not only useful to support specialized learners, but can also be used to reduce the overall communication cost of permissioned blockchains and to improve their storage scalability.
@InProceedings{hellings_et_al:LIPIcs.ICDT.2020.17, author = {Hellings, Jelle and Sadoghi, Mohammad}, title = {{Coordination-Free Byzantine Replication with Minimal Communication Costs}}, booktitle = {23rd International Conference on Database Theory (ICDT 2020)}, pages = {17:1--17:20}, series = {Leibniz International Proceedings in Informatics (LIPIcs)}, ISBN = {978-3-95977-139-9}, ISSN = {1868-8969}, year = {2020}, volume = {155}, editor = {Lutz, Carsten and Jung, Jean Christoph}, publisher = {Schloss Dagstuhl -- Leibniz-Zentrum f{\"u}r Informatik}, address = {Dagstuhl, Germany}, URL = {https://drops.dagstuhl.de/entities/document/10.4230/LIPIcs.ICDT.2020.17}, URN = {urn:nbn:de:0030-drops-119418}, doi = {10.4230/LIPIcs.ICDT.2020.17}, annote = {Keywords: Byzantine learner, coordination-free checkpoint protocol, delayed-replication, information dispersal, consensus} }
Feedback for Dagstuhl Publishing