Dynamic Replication Scheduling for Cloud Datacenters Based On Workload Statistics

Cloud providers often achieve high service reliability and availability by using replication technologies. Typical replication mechanisms usually adopt fixed schedules for replicators by transferring data to their replicas at a specified time. While replications and applications are performed simultaneously, they may cause resource contention and result in replication failures. Once a replication fails, its replicator will retry the replication to extend the transmission time and cause additional CPU overhead in the system. As a result, the response time of the replication is increased. In this paper, we present a mechanism of dynamic replication scheduling based on workload statistics (DRSWS) to dynamically adjust the size of each replication batch according to the workload statistics. The experimental results show that DRSWS can efficiently alleviate system overhead and reduce response time.