I am looking for information regarding the performance of a restore using cumulative incremental vs differential incremental backups. I understand the difference between the two. If the full and incremental backups are to a disk based backup system, i.e. a 5230 netbackup appliance, which full and incremental backup combination is better for a restore, cumulative or differential? I understand in a tape environment that a restore using a cumulative backup will reduce tape mounts but in a disk environment where no tape mount is required, is there a performance improvement when restoring a server using a full and a cumulative incremental backup? If so, how much better is it over a full and differential backup? If anyone can share performance data, it would be appreciated. I am assuming a weekly full and daily incremental backups.
No performance comparison data here. However, I would consider the fundamentals in making the decision. (You know this but others searching something similar might benefit) - Differential Incremental (DI) means that you restore from a FULL backup and then all subsequent DI backups till the target date. Hence, the possibility of multiple tape mounts. Cummulative Incremental (CI) means a restore from FULL backup and then from the target date CI backup. So, basically around two mounts of tapes (or more if there image spans multiple tapes).
Disk based images reduce the mount time and positioning time that tapes require, however, if there is a lot of daily change in your data, the CI backups themselves take a long time to finish as opposed to DI backup.
You need to consider the type of data too. Is deduplication involved in the disk storage unit? If yes, data with less deduplication rate will rapidly fill up your disk. Sending that data to tape might be more logical (e.g. encrypted data, video files, etc).
So, it depends on your environment and RTO requirements.
PS: I usually go for weekly CI when I'm doing monthly FULLs for large volumes.
If you are backing up a database or database application such as Microsoft Exchange or SharePoint, consider that taking cumulative incremental means that the database can't truncate the transaction logs (delete the log files) until the next full backup. I think Exchange users tend to prefer incremental backups because they can have tens of thousands of 1 MB log files generated per week.
Accelerator does change the picture at least for Exchange. Exchange itself is slow at replaying the logs from an incremental backup. With Accelerator, you may as well take only full backups. Full Accelerator backups take the space of differential backups but restore in a single step.
here's hoping that someone more knowledgable might comment on...
...how TIR "catch-up processing" for a restore might make either or both of a performance difference and/or an elapsed duration difference between a "full + one cinc" restore versus a "full + multiple diff" restore.
That is good to know for Exchange. For other plain, vanilla server backups and accelerator, If I am reading information regarding Accelerato correctly, any incremental backup creates a new full backup for restore. Therefore, it doesn't matter which you choose to perform, dufferential or cumulative, provided you have a good base full backup. Since cumulative incrementals do not reset the archive bit, you will be backing up more data daily if the data is changing. If you choose differential incrementals, the archive bit is reset and you are only backing up the changes made since the last backup. If you have a lot of changed data and you are running daily incrementals, the cinc would take longer as the days accumulate until you do another full backup. Whereas, a dinc would not take as long daily. The end result of either incremental backup would result in a new full to be used for recovery. Am I stating this correctly?