I thought some of our file servers were rather large, but you have us beat for sure! We have lots file servers in the 5 to 15 TB range. They are all VMware guests and backed up using SAN transfer. The largest can take anywhere from 10 to 18 hours to run a full backup. The incremental backups are quick since they use the VMware change block tracking. One thing to watch out for is when an VM admin adds space to one of the disks in vCenter, it re-sets the change block tracking for the VM so it ends up forcing the backup to be a full. (commvault will run an incremental, but the size is a full)
From a CV licensing perspective this is great.
We are able to still run file level restores from these backups it takes a little longer than a traditional file level restores but not too bad.
For a full server restore, the Full VM will be faster(SAN mode transfer) than a file level restore.
Are you looking to put the full petabyte on one server? That would be scary if the server runs into any issues... server patches and what not. I would try to spread it across a few servers.
We do also use a Netapp for some file sharing. It is set to replicate to another Netapp at our DR site. The Commvault backups against it are super slow. Luckily these files are a write once read many situations. So, the data once written does not change. So we are able to run one time backups of the older volumes and keep them in Commvault and they are good going forward.
I think some folks would use the snapshots on the arrays to keep the backups. I still subscribe to the 3-2-1 backup strategy.
Cohesity and Rubrik look like interesting options for large fileservers. We tried out Cohesity for a specific purpose, but it was not able to support the speeds we needed. For normal user usage it would have been ok.
Hopefully this helps a little bit.