Data backup copies all or some data sets of an application from a host disk or storage array to another storage medium to prevent misoperations or failures from causing data loss. CAS CVM relies on array-based replication and continuous data protection (CDP) to back up data.
Array-based replication can be performed synchronously or asynchronously.
Synchronous replication ensures data consistency between the storage volumes in a replication pair. In synchronous replication, each IO operation releases resources only after both the local and remote volumes in a replication pair return write operation completion. Synchronous replication provides the highest level of data integrity at the cost of decreased performance caused by data transmission latency, and it requires the round trip delay between source and destination arrays to be short. Typically, synchronous replication is used for short-distance replication (10 to 100 km, or 6.21 to 62.13 miles) in scenarios that require strict data consistency and near-zero data loss, such as internal systems of banks.
Figure-1 Synchronous replication
Asynchronous replication is performed periodically and thus cannot ensure data consistency between volumes in a replication pair. In asynchronous replication, a local volume creates a snapshot after it finishes a write operation and copies the snapshot to a remote volume. Asynchronous replication offers high performance but does not guarantee zero data loss because source and destination volumes might have inconsistent data. Asynchronous replication does not require high bandwidth or short transmission distance, which makes it suitable for systems that require high performance, have light write loads, and does not require high array IOPS performance or short delay, such as databases and file systems.
Figure-2 Asynchronous replication
CDP backs up production data on a system to a target repository every time a change is made. By capturing and tracking data modifications, CDP maintains a continuous journal of data changes and makes it possible to restore a system to any previous point in time. CDP depends on agent software to monitor IO changes.