Geo-Replicated Storage with Scalable Deferred Update Replication

This extended abstract discusses scalable deferred update replication in geographically distributed settings. SDUR scales deferred update replication, a well-established proach used in several database replicated systems, by means of data partitioning. SDUR distinguishes between fast local transactions and slower global transactions. Although local transactions scale linearly with the number of partitions (under certain workloads), when deployed in a geographically distributed environment they may be significantly delayed by the much slower global transactions-in some settings global transactions can slow down local transactions by factor of 10. We presented two techniques that account for this limitation: Transaction delaying is simple, however, produces limited improvements; reordering, a more sophisticated approach, provides considerable reduction in latency of local transactions, mainly in deployments where global transactions harm local transactions the most.