Yet another forward and reverse incremental post...

[sgravel]

OK, so I have read the user guide and I feel that I understand how these two compare and contrast with one another, but I have some questions about how to set up Veeam to mimic the current RPO we have right now with our old DPM solution.

We're vSphere. We have maybe 30 VMs on 3 hosts that we are concerning ourselves with. At present, we are backing them up with DPM agents in the guest OS, getting recovery points every 30 minutes, only during production hours. Once each evening, a full backup is synthesized from the recovery points. At this point, Exchange and SQL logs are committed and flushed. The half hour granular restores remain available on disk for two weeks. The next morning, the latest full backup is copied to tape. It's retained for two weeks. We also have monthly tapes that are kept for one year, and yearly tapes that are kept for eight years.

Right off, I know that tape is 3rd party at this point, and that's fine. I will probably keep DPM around for a while, and back up the Veeam server's backup repository as a DPM client. Either that, or install Nova Backup on the physical backup repository and attach a local tape drive, so that once DPM cycles off the radar, I can wind it down.

Where I start to get confused is that forward incrementals are billed as "D-D-T friendly", with the thought that less will need to be written to tape each day (i.e. only the incremental files). I recognize that it is a little over the top, but at least for now, we are required to keep daily tapes as a safety net, and they each contain that day's most recent COMPLETE backup. Not incrementals. I can grab one tape and know that I can do a restore.

Based on that requirement, I would think that reverse incrementals start looking pretty good, despite the tape requirements. Set the 3rd party tape job to grab the .vbk files in the backup repository, and you have what you need - the latest complete, full backup. Am I thinking about this correctly? I know that reverse is 3x more I/O intensive. Is that likely to be an issue given my 30 minute RPO? We're going with enterprise class SATA DAS for the backup repository. I also wonder, given that we are running the tapes during the day time, the fact that the vbk contents will be changing (reverse) during the tape job's window will present any issues. Maybe it would make more sense to move the tape runs to overnight when there is less going on. That would be OK.

If the increased I/O load of reverse is going to push us toward forward incrementals, and we choose to create daily synthetic fulls, do those get created at a set time, or do they just happen on the first run as scheduled on the specified day? Also, if I go this route, and try to get around the increased storage requirements, it sounds like I can only convert the first preceding chain of incrementals to reverse incrementals, so I will still have a considerable number of synthetic fulls to keep around if I want to go back two weeks on disk.

A third option may be to run two kinds of jobs against the VMs. A forward incremental with weekly synthetics for my 30 minute RPO / 14 day retention (less I/O than a reverse, only 2-3 syn. fulls), and another forward incremental with 24 hour RPO and daily synthetics that is only retained for one day. Even so, that's 3 or 4 full backups worth of space to keep around, plus incrementals.

My head is sort of spinning - what would you do?

Also, I do realize that anywhere I said I want to keep stuff for X days, I need to do some math and figure out how many RP's that is.

Thanks!

[Vitaliy S.]

Based on that requirement, I would think that reverse incrementals start looking pretty good, despite the tape requirements. Set the 3rd party tape job to grab the .vbk files in the backup repository, and you have what you need - the latest complete, full backup. Am I thinking about this correctly?

Yes, this is the preferred backup mode, given that you have to offload the full VM backup every time.

I know that reverse is 3x more I/O intensive. Is that likely to be an issue given my 30 minute RPO? We're going with enterprise class SATA DAS for the backup repository.

Cannot say for sure, as the answer depends on the performance of your target repository and the amount of VM data you're going to backup, but I believe you should be fine. You need to give this approach a try

]I also wonder, given that we are running the tapes during the day time, the fact that the vbk contents will be changing (reverse) during the tape job's window will present any issues. Maybe it would make more sense to move the tape runs to overnight when there is less going on.

If you're going to run your backup job every 30 minutes, then this time period might not be enough to offload the entire VBK file to tape, so I would suggest using your night time for this activity.

If the increased I/O load of reverse is going to push us toward forward incrementals, and we choose to create daily synthetic fulls, do those get created at a set time, or do they just happen on the first run as scheduled on the specified day?

Synthetic full is triggered right after successful incremental job run which should happen on the same day the synthetic full is scheduled on.

Also, if I go this route, and try to get around the increased storage requirements, it sounds like I can only convert the first preceding chain of incrementals to reverse incrementals, so I will still have a considerable number of synthetic fulls to keep around if I want to go back two weeks on disk.

Correct.

[sgravel]

Thanks for the answers. We have not purchased the production hardware yet, so we may see if we can squeeze in SAS over SATA, increase the drive count so that we can run RAID 0+1, or maybe both to increase the performace on the backup repository so that the reverse incrementals are easier to handle.