Backup Repository - ISCSI to ESXi or directly to VM

[vlad_diverse]

I have seen a lot of posts, like this. But, would like an advise in year 2021.
I have Lenovo DE6000 and ESXi host. I need to create a large ~ 1PB backup repository. I have several ways to connect them, but because of costs they are connected directly 8x10GBps LAN. Cable from one adapter on storage to another one on server. Eight 10GBps connections.
Not going to configure VMFS datastore, but configured ISCSI all 8 connections to ESXi server.I see traffic on all 8 ports, means multipathing.
Then mounted RAW LUN to Windows VM, witch will be a Veeam Repository and formatted as ReFS.
I could connect directly to Windows VM iSCSI initiator, bypassing ESXi ISCSI configuration, but my argument that if I loose Repository VM, I can connect RAW disk quickly to a different Windows VM on that ESXi host, bypassing ISCSI configuration in Windows VM.
Any suggestions?

[PetrM]

Hi Vladimir,

Basically, I'd use physical repository to separate the production environment from the backup storage itself. One more downside of virtual repository is that you'll need to make more manual steps to reconnect the storage to another server in case of ESXi failure. If you're seeking for general best practices, then you may refer to this page of our best practices guide and to that one about block-level repositories.

Regarding your initial question, I believe that you can connect the storage via ESXi iSCSI if it does not create performance bottlenecks for backup jobs. By the way, I recommend to consider some additional storage to follow 3-2-1 rule.

Thanks!

[vlad_diverse]

Thanks Petr, I see now, that physical repository is a recommended way.
As a storage person, I do not see on how ESXi ISCSI stack will create any bottlenecks with 80GBps of bandwidth and RAW disk.
But I see, that go physical has its benefits in raw performance.
I was thinking about creating an ESXi Cluster in the future, where you can connect multiple storage devices via ISCSI and VM repositories could shift between nodes.
It could create additional flexibility for hosted backups, where you can share infrastructure between clients.