Single Box Local Backups - Passthrough, VHDX, or Host as Repository?

[c.haydock]

The Scenario

• Two "production" hypervisors with shared nothing SSD storage replicating VMs between them
• One "backup" hypervisor with large spinning HDD storage running a Veeam Backup Server as a guest VM

General Q&A

• Why the backup hypervisor rather than bare metal? - It is mainly so that in the worst of worst cases (all be it unlikely I've unfortunately seen it happen), both production hypervisors could die and... although it would run as slow as molasses in January... a recovery of the production workloads could be accomplished on the backup hypervisor.
• What about 3-2-1-1-0? - It's all being addressed, but if if you see a particular impact as it pertains to my question below, feel free to bring it up.

My Question
As it pertains to the storage of backups I see a few options and I'm interested in what would be "best practice" as well as the reasons why. At the very least what if any pros/cons of the following are...

1. create a large VHDX file/drive and attach it directly to the Backup Server VM to use as a local repository.
2. Use Hyper-V's passthrough capability to attach the underlying storage disks directly to the Backup Server VM to use as a local repository.
3. Register the host hypervisor as an external repository from the Backup Server VM leveraging the data mover.

My Take So Far

1. PROs
   • SIMPLICITY! - Treat it like you would most any other VM
   • It allows for fairly easy migration of backups to another host without loosing the benefits of ReFS block cloning (no re-hydration penalty when moving)
   CONs
   • It could be a significant performance hit during restorations (not sure on this... just guessing)
   • Once the VHDX grows, to my understanding, you can't shrink an ReFS volume. So I'll have to mind my P's and Q's for space being used
2. PROs
   • It may be slightly better on performance
   CONs
   • Migrating backups means either dealing with re-hydration penalties or moving physical disks to a new host
3. PROs
   • I expect disk performance would match that of passthrough... perhaps even better?
   CONs
   • Licensing? I believe the verbiage from MS is something to the effect that to use the license for two guest VMs the host can take on no other roles other than being a hypervisor?
   • CPU resources in this configuration may be in contention with the Backup Server guest VM thus hampering performance for backups and recoveries (the backup server has a fairly low core count)
   • Migrating backups means either dealing with re-hydration penalties or moving physical disks to a new host

I don't feel that any one in particular is better than the other, although I feel inclined to go with #1. I also don't feel like there is anything so grossly wrong with any of them that would preclude implementing any one of them (other than maybe being hit with extra MS licensing in #3). What are your thoughts??? Many thanks in advance for your input/feedback

[ejenner]

I'd probably go for 2 as it seems more straight forward and by the sounds of it the backup files would be easier to access as they're stored directly on physical disks?

[nmdange]

If you're doing ReFS, storing the data in a VHDX makes it a lot easier to move it to new hardware without losing the space savings from block cloning.