Skip to main content
Enterprise applications

Oracle database backup tiering

Contributors jfsinmsp

Traditional application backups include products such as Oracle Recovery Manager, which create file-based backups outside the location of the original database.

`tiering-minimum-cooling-days` policy of a few days preserves the most recent backups, and therefore the backups most likely to be required for an urgent recovery situation, on the performance tier. The data blocks of the older files are then moved to the capacity tier.

The auto policy is the most appropriate policy for backup data. This ensures prompt tiering when the cooling threshold has been reached irrespective of whether the files have been deleted or continue to exist in the primary file system. Storing all the potentially required files in a single location in the active file system also simplifies management. There is no reason to search through snapshots to locate a file that needs to be restored.

The snapshot-only policy could be made to work, but that policy only applies to blocks that are no longer in the active file system. Therefore, files on an NFS or SMB share must be deleted first before the data can be tiered.

This policy would be even less efficient with a LUN configuration because deletion of a file from a LUN only removes the file references from the file system metadata. The actual blocks on the LUNs remain in place until they are overwritten. This situation can create a lengthy delay between the time a file is deleted and the time that the blocks are overwritten and become candidates for tiering. There is some benefit to moving the snapshot-only blocks to the capacity tier, but, overall, FabricPool management of backup data works best with the auto policy.

Note This approach helps users manage the space required for backups more efficiently, but FabricPool itself is not a backup technology. Tiering backup files to object store simplifies management because the files are still visible on the original storage system, but the data blocks in the object store destination depend on the original storage system. If the source volume is lost, the object store data is no longer useable.