Which is better VMWare pass-through RDM vs Direct Attach storage

[iamloki]

Hi Everyone,

I’m seeking advice on the better option for achieving optimal performance for backups. Below is the current setup and planned changes:

Current Setup:

vSphere backups are being performed using 1 virtual VBR and 2 virtual Proxies.

Backups are currently written to a Quantum DXi 4700, but performance is very slow.

Planned Changes:
We are replacing the Quantum DXi 4700 with an HP MSA 2040, formatted with ReFS for use as a Windows Repository.

Options Being Considered:
Option 1: Present the MSA 2040 as a pass-through RDM to a virtual Windows Repository (hosted on the ESXi server).
Option 2: Directly connect the MSA 2040 to a dedicated physical Windows Server, with the storage formatted as ReFS.

Question:
Which option is likely to provide better performance in this scenario, and are there any other considerations (compatibility, or potential bottlenecks) that I should take into account?

[Andreas Neufert]

I would say it is pretty similar and depends more on the connection throughput.
But in general I would spin it a bit differently.

In the face of the current ransomware thread, I would connect the MSA 2040 to a physical server and use the Hardened Linux Repository ISO from Veeam to run a Hardened Repository on the physical server using the MSA as space. If you did not bought already the MSA2040, check out the option to buy an HPE Apollo system or if need more space 2 systems in a Veeam Scale out Backup Repository (SOBR). With the MSA2040 you have an additional attack vector the management of the system while with the Apollo you would just have the system itself. Please check out this entry that lists exactly the system, the bill of material and the configuration/implementation guide: https://www.veeam.com/sys1140

[iamloki]

We have already purchased MSA2040, we somehow have to make it work and I'm keen to know the best approach.

[Andreas Neufert]

- Connect the MSA2040 (directly if possible to avoid another attack vector) to a physical server. As you wrote with FC, then use 2 HBAs within the server (not dual port).
- Isolate the management of the MSA2040 as much as possible. At best disconnect management ports or add another small system just for management that is not connected to the regular network, just to the MSA management. Like a raspberry pie or old window notebook or so.
- Install Veeam Hardened Linux Repository ISO on the server.

This setup will give you immutability.

Do not forget to create backup copies, for example to the Veeam Cloud Vault. Main point is to have a backup copy at another place.

[Origin 2000]

+1 for the phys. Repo!

Consider to use Veeam Hardenet Linux with XFS instead of Windows with ReFS.

Regards,
Joerg

[Spex]

I wouldn't use the msa 2040 in a new setup since its total outdated - since 2023 end of support.
See https://support.hpe.com/hpesc/public/do ... cale=en_US
I also would go to Veeam hardened Linux with XFS instead of Windows ReFS.
With ReFS we had some things to learn, XFS is rock solid.
Prio1 is security and reliability/ Prio 2 maybe performance

My 2 cent

[MVanek]

+1 for dedicated physical repo (hardened one .. I dont see much of a problem with MSA storage as a repo. Yes 2040 is a bit of the old HW.
We are not aware of your detailed backup scenario, budget etc.... Generally technically speaking MSA2040 is good piece of storage. If we speak about Apollo - nice machine at all, but price of this beast might be very different comparing to MSA.
Anyway - there is one rule I would follow generally (just repeating what Andreas already said) - Keep in mind that storing your backups on virtual machine (even with connected dedicated physical storage) dramatically increases your RTO in case of HW failure of that production HW.
Try to store your backups as simple as possible and as isolated as possible, also try to have as much copies as possible.
Unfortunatelly I see a lot of companies saves money when it comes to backup or generally non-prod solutions. Anyway even with that aproach there are some ways to do it with less pain after all
If there is a bunch of old servers/storage or even NAS (I hope not at all ). One can recycle these machines and use them as backup components.
With that scenario is more then ever very important to have multiple (same) copies of backup data.
In case you lose your production environment - these machines can save your day. Even the old ones
I can imagine that MSA 2040 can do much of music when it comes to instant vm recovery. All you just need is some piece of ESXi to run those IVMR VMs on AND a Backup Server/Mount Server with valid backups somewhere --- And this is the culprit - if this guys were VMs on that crashed ESXi host, you will spend some time with installing / configuring blah blah of those.
So divide these two worlds - production for production and backup for backup and RESTORE.