Multi-site Gargamel: optimistic synchronization for reliable geo-replicated databases

Databases scale poorly in distributed configurations. This is mainly due to the cost of concurrency control and to resource contention. The alternative of centralizing writes works well only for read-intensive workloads, whereas weakening transactional properties is problematic for application developers. In a previous work we introduced Gargamel, a middleware that spreads non-conflicting update transactions to different replicas, but still provides strong transactional guarantees. We extended Gargamel to geo-replication settings. If a data-center fails, the database remains available at other locations. We minimize the synchronization cost, synchronizing optimistically replicas at distant data-centers. The evaluations of our prototype using distant Amazon data-centers show that Gargamel improves both response time and load by an order of magnitude when contention is high (highly loaded system with bounded resources), and that in the geo-replicated case, the slow-down is negligible.