File System RPO Snapshots in Near Real-Time with Asynchronous Replication

Abstract

In the File System level Disaster Recovery configurations, asynchronous data replication is often used between the local Primary File System and the remote Recovery File System to avoid latency to the applications running on the local Primary File System. The individual data updates of the local Primary File System are replicated to the remote Recovery File System in the background asynchronously after applying some delay called asynchronous delay. During the asynchronous delay, optimization methods like coalescing smaller contiguous write operations into a more extensive write operation and eliminating short-lived data updates are applied to the data updates to reduce the network bandwidth requirement for data replication. However, the asynchronous delay likely causes the delay in taking periodic Recovery Point Objective (RPO) snapshots on the Recovery File System for data consistency because of the large amount of data pending replication to the remote Recovery File System before taking RPO snapshots. This delay in taking RPO snapshots could cause more data loss, causing RPO violations if disaster hits the local Primary File System. Taking RPO snapshots strictly at RPO intervals is critical. This paper describes a new efficient procedure for taking RPO snapshots close to the RPO interval without delay by replicating pending data updates to the remote Recovery File System earlier without waiting for the asynchronous delay. Based on pending data replication, the aggregated network bandwidth between the local primary location and the remote recovery location, and the aggregated rate of data generated by applications, the early replication time before the next RPO time without waiting for the asynchronous delay is calculated.