blazing salmon
Hi everyone, I have a query about a Hyper-V environment with SQL and Exchange. Is it more efficient to connect many .vhdx files (one for the OS, another for the MDF databases, and another for the LDF files)?
Should each .vhdx be placed on a separate volume and LUN?
And should each LUN be connected via iSCSI to the Hyper-V server (creating a Cluster Storage)?
Or is it more efficient to use iSCSI and LUNs within the Windows VM instead of using .vhdx for the MDF and LDF files?
The goal is to use Veeam to replicate Hyper-V VMs from a Hyper-V cluster + NetApp AFF90c in the primary data center to a backup data center, also with Hyper-V cluster servers + a NetApp ASA150.
Cluster nodes use iSCSI to access LUNs as Cluster Storage, solely for accessing 1 .vhdx with the Windows VM OS.
Inside the VM, I have new iSCSI connections to other LUNs with the MDF and LDF files.
Veeam copies the virtual machine but does it copy the LUNs as well? Or does it not?
Should I replicate those LUNs with SnapMirror and in case of a DR, first activate the replicated SVM, verify that they have the same LIFs, bring them online, and then start the replicated VMs so they see the same SVM (iQN)?
low wedge
If you're attaching iSCSI LUNs inside the Windows VM, Veeam will not replicate those as they are not virtual hard drives in the Hyper-V Config for the VM. You'd need to present LUNs from ONTAP to the Hyper-V hosts and then create the disks as VHDs.
Veeam doesn't have, I believe, storage integration with ONTAP when using Hyper-V, so Veeam will be doing all the data moving and not leveraging SnapMirror. I wish they'd add that support like they have for VMware.
Regarding the VHDX per LUN or not, I'd personally go with more LUNs on the A90. ONTAP thrives with more parallel access; more FlexVols.
ONTAP 9.15.1 (which you'd have to at least be on to support an A90) does have some improvements to increase resources available to single FlexVols, but more volumes and spreading the load is typically more performant.
I always start with at least 8 FlexVols, 4 per controller, and then a LUN per volume (thats on smaller A250 systems, for example).
For an A90 I'd go 8 or 12 volumes per-controller initially as ONTAP loves those CPU cores. How you want to distribute your VHDX's across those LUNs is up to you. But if you have VMs that hit all their disks at the same time, maybe try and spread those VHDX's across multiple volumes and both controllers.
blazing salmon
Hi, thanks for the response.
Today I have 1 volume + 1 LUN for each database file and for each database, I have many volumes and LUNs.
And the databases are spread across the controllers, for example, a database has the MDF on one volume+LUN on one controller and the LDF on another volume+LUN on another controller.
All volumes+LUNs that are data (SQL or Exchange) are connected via iSCSI from each Windows VM directly.
On each Hyper-V host in the cluster, it does iSCSI to LUNs where the VHDX of each VM that has the VM OS installed is located.
The question is if I connect all the LUNs via iSCSI on the Hyper-V hosts, move all the MDFs and LDFs to VHDX files and present the VHDX files to each VM so that Veeam can replicate the entire SQL VM to the other data center.