Restore points vs. Days for Hourlies

[chrstn32]

I'm sorry for the amount of setup I need to explain what I'm asking. I don't know how to make this shorter.

I have an hourly backup job that runs for the 4 most important VMs in our production environment. I originally had it set up to do Forever Forward Incremental with no fulls. I had retention set up for 72 restore points (the idea being that I can increment back to any hour on any give day for the last 3 days). However, it started taking over an hour for the merge to happen, and of course it was happening at the end of every job run. This was keeping the job from running again hourly, and I was only getting every other hour.

After talking to a Veeam engineer on the phone, they recommended that I run the job with periodic synthetic fulls. That way, it will not try to merge the files every hour. This works, of course. However, now it is keeping closer to 120 restore points, instead of just 72, because the period synthetic full runs on Saturday. It took about 6 hours to run this past Saturday, which is okay mostly, because we do not have a lot of customer use over the weekend.

My question (finally) is, are we doing this right? Is there a more efficient way? Also, what will happen if I change the retention to 3 days instead of 72 restore points, will that hurt or help the amount of incrementals it keeps? Also, if I set it to create period fulls every day or every other day instead of just on Saturday, is there any way I can be sure that those won't run between our heaviest production hours (like 7AM to 7PM M-F)?

Sorry for the novel size question from a relative Veeam Noob. Any help is greatly appreciated! Thanks!

[Mildur]

to 3 days instead of 72 restore points, will that hurt or help the amount of incrementals it keeps? Also, if I set it to create period fulls every day or every

With weekly synth fulls and hourly backups, you will always have this 24*7 restorepoints. Setting it to 3 days doesn‘t change anything.
Veeam needs all 24*7 restorepoints to restore data from that chain. A chain is a Fullbackup and all followed incremental files until a new Fullbackup (active or synthetic full) is generated.
Second, veeam will keep a second backup chain until the new chain is three days old. You will have between 168 and 336 restore points all the time. This is how forward incremental works.

Can you tell us more about your backup repo type?
Eventually you can optimize the synthetic full times if you use FAST Clone on a supported Backup repository.


Or you could configure a second synthetic full sometimes midweek, this will help you to get smaller backup chain length. Smaller chains and a retention of three days could get you to between 72 and 144 restorepoints.

[chrstn32]

So, we have a relatively small environment at this location compared to what a lot of users will have. This backup repo is an NFS share and is added to the VBR server as NAS. I would rather keep less than 336 restore points if possible, more for simplicities sake than anything else. We are pretty okay with storage so far, and I can most likely expand storage if it came down to it. This environment will never see any other type of storage though, it will always just get NFS. Against my initial recommendations, the management chose not purchase a real storage solution for this particular environment. (Thankfully, the VMs themselves are running on a NetApp flash storage system for their datastores, not an NFS). Is it weird/bad to have as many as 336 restore points?

I'm just looking for the most efficient way to do this that actually works. To add SLIGHTLY more background to this story. I have a script that automatically initiates an instant restore on any of these VMs that throw a "heartbeat is missing" alarm through Veeam ONE. 10 minutes after that alarm goes off, the script runs and attempts to ping the VM... if unable to ping it, an instant restore takes place.

What would happen if I set the job to take the synthetic full (or even an active full) every day or every other day, instead of once a week? When would those merges happen or how could I make sure they only happen at night?

I'm also considering replacing this hourly job with CDP now that we've upgraded to v11... I just haven't been able to get the I/O Filters deployed and test out how this will work. Unfortunately it's hard to get any downtime on the system to implement that right now.

[Mildur]

I recommend to keep the restore point chain small.
If one vib file is missing, all restore points after the missing vib file are garbage. But with a NFS Repo, it should‘t be a issue to have a chain with dozens of incrementals. With dedup appliances, this would be more of an issue.
Taking a synthetic Full every day would result in 24 restore point backup chain. With a retention of three days, you would get 4x chains with each 24 restore points on your backup repo. The first one of each day would be the synthesized full at 00:00.
If you have the storage for 4 Fullbackups and their incrementals, test it.

Another thing you could try, is to move to active full instead of synthetic fulls. You have your production on NetApp Flash Storage. Taking an active Full should‘t take to long, if your Important vms are not in the Terabyte size.
If the backup NAS itself is low on performance or connected with a low bandwith, then active full instead of synthetic full can help to speed up the process.

I'm also considering replacing this hourly job with CDP now that we've upgraded to v11

If you need application consistency with this important vms, a normal replication job would be better for your case.
With only 1 hour RPO, a veeam replica job should be used. The only thing is, that you can configure only 28 replica points. This is a vsphere limitation.

[chrstn32]

Thank you so much for all the good information. I have a much clearer understanding of what's going on. The settings we'll try out for now is a Synthetic Full that runs every day and 3 days of retention.

I'll let that run for a week or so until the chains are stabilized and I'll report back here with what we see with it. I think it should work out pretty well.

I was actually using a replica job originally, but a Veeam rep that I talked with actually recommended using regular backups and setting the script up to run an instant restore instead of using a replica job. I can't remember his reasoning off the top of my head...

[chrstn32]

Just wanted to come back and post a final update on this.

Running this job as specified in my last post is working as expected and is a good mix of performance vs. storage. The backup happens hourly and runs for less than 10 minutes, then every day around midnight a synthetic full is produced.

I appreciate the assistance from @Mildur.

[Mildur]

Glad it worked, Chris