Migrating a 10TB VM from ESXi to PVE/Ceph

Migrating a 10TB VM from ESXi to PVE/Ceph

[Fabien Sirjean]

Dear all,

After successfully migrating most of our infrastructure (300+ VMs) from 
VMware ESXi to PVE, we're struggling with one last critical VM: our 
BlueMind/Cyrus mail server (~800 users), with a 10TB virtual disk.

Our setup is:

Source: ESXi → Dell EMC VPLEX (FC) → Dell Unity
Target: Proxmox VE → Ceph RBD (dedicated cluster)

We tried the Proxmox ESXi importer, but I/O performance degraded so much 
during the live import that the mail service could not sustain its 
normal workload.
We then stopped the service to complete the migration, but the import 
eventually failed after almost 3 days, forcing us to start over.

Application-level migration/synchronization is not an option, so we're 
looking for a block/storage-level approach with minimal downtime.

Since the datastore ultimately resides on our FC SAN, could we bypass 
ESXi entirely and temporarily present the relevant VPLEX/Unity LUN to a 
PVE or migration host, mount/access the VMFS datastore read-only, and 
copy/convert the VMDK directly to Ceph RBD?
Even better, is there a practical way to perform the bulk copy while the 
VM is running, then only synchronize the changed blocks during the final 
cutover?

Any advice or experience with direct VMFS access, VPLEX/Unity 
snapshots/clones, or other approaches for this kind of migration would 
be greatly appreciated.

Thanks!

Re: Migrating a 10TB VM from ESXi to PVE/Ceph

[Adam Thompson]

I migrated some very large (not quite as big as yours, only ~3TB) VMs by pivoting through a common NFS server.

Vmotion the VM to NFS.
Run the ESXi importer without any disks.
With the ESX VM shut down, manually attach the VMDKs from NFS.
Boot the VM under PVE.
Finally "move storage" in PVE.

Major downside: you need 2x or 3x the disk space.

I also used an offline variant of that with one PVE node acting as the NFS server, then "qm importdisk" locally on that node.

-Adam

Get Outlook for Android<https://aka.ms/AAb9ysg>
________________________________
From: Fabien Sirjean <fsirjean@eddie.fdn.fr>
Sent: Monday, 07 September 2026 02:08:23
To: pve-user@lists.proxmox.com <pve-user@lists.proxmox.com>
Subject: Migrating a 10TB VM from ESXi to PVE/Ceph

Dear all,

After successfully migrating most of our infrastructure (300+ VMs) from
VMware ESXi to PVE, we're struggling with one last critical VM: our
BlueMind/Cyrus mail server (~800 users), with a 10TB virtual disk.

Our setup is:

Source: ESXi → Dell EMC VPLEX (FC) → Dell Unity
Target: Proxmox VE → Ceph RBD (dedicated cluster)

We tried the Proxmox ESXi importer, but I/O performance degraded so much
during the live import that the mail service could not sustain its
normal workload.
We then stopped the service to complete the migration, but the import
eventually failed after almost 3 days, forcing us to start over.

Application-level migration/synchronization is not an option, so we're
looking for a block/storage-level approach with minimal downtime.

Since the datastore ultimately resides on our FC SAN, could we bypass
ESXi entirely and temporarily present the relevant VPLEX/Unity LUN to a
PVE or migration host, mount/access the VMFS datastore read-only, and
copy/convert the VMDK directly to Ceph RBD?
Even better, is there a practical way to perform the bulk copy while the
VM is running, then only synchronize the changed blocks during the final
cutover?

Any advice or experience with direct VMFS access, VPLEX/Unity
snapshots/clones, or other approaches for this kind of migration would
be greatly appreciated.

Thanks!

Re: Migrating a 10TB VM from ESXi to PVE/Ceph

[Tom Weber]

Hi,

I'd try to make your FC SAN Storage visible to your PVE Cluster - so that it can "see" the .vmdk Files.

Next define a VM with the needed Hardware Settings (CPU, RAM, Network...) but small / empty Disks on your FC SAN Storage - choose VMDK Format.

This should give you a machine description in PVE (in /etc/pve/nodes/<NODE>/qemu-server/<ID>.conf) and vm-<ID>-disk-<x>.vmdk files on the FC SAN Storage.

Now with the VM shut down (in both worlds), you could rename the production VMDK Files to the vm-<ID>-disk-<x>.vmdk names used in the <ID>.conf file and you should be able to fire up the VM on the PVE cluster working with the vmware disks on the FC SAN (current Data).

Once the VM is running you could do a live migration (move Storage) of the VMDK disk to Ceph or whatever other PVE Storage.


 From time to time I did something similar and I dont see why it shouldn't work but please test and experiment with smaller VMs first - and if possible work with Snaphots/Backups whatever before mounting the production .vmdk files in PVE.

Regards,
   Tom


Am 07.09.26 um 09:08 schrieb Fabien Sirjean:
> Dear all,
> 
> After successfully migrating most of our infrastructure (300+ VMs) from VMware ESXi to PVE, we're struggling with one last critical VM: our BlueMind/Cyrus mail server (~800 users), with a 10TB virtual disk.
> 
> Our setup is:
> 
> Source: ESXi → Dell EMC VPLEX (FC) → Dell Unity
> Target: Proxmox VE → Ceph RBD (dedicated cluster)
> 
> We tried the Proxmox ESXi importer, but I/O performance degraded so much during the live import that the mail service could not sustain its normal workload.
> We then stopped the service to complete the migration, but the import eventually failed after almost 3 days, forcing us to start over.
> 
> Application-level migration/synchronization is not an option, so we're looking for a block/storage-level approach with minimal downtime.
> 
> Since the datastore ultimately resides on our FC SAN, could we bypass ESXi entirely and temporarily present the relevant VPLEX/Unity LUN to a PVE or migration host, mount/access the VMFS datastore read-only, and copy/convert the VMDK directly to Ceph RBD?
> Even better, is there a practical way to perform the bulk copy while the VM is running, then only synchronize the changed blocks during the final cutover?
> 
> Any advice or experience with direct VMFS access, VPLEX/Unity snapshots/clones, or other approaches for this kind of migration would be greatly appreciated.
> 
> Thanks!

RE: Migrating a 10TB VM from ESXi to PVE/Ceph

[Adam Thompson]

I went with NFS precisely because FC storage isn't generally something PVE can mount - it's block storage, so it'll have (most likely) a VMFS5 filesystem and raw files with extents inside it, not VMDKs.  Pivoting through an NFS server flattens everything into a VMDK, which PVE can then mount and see and use.
-Adam

-----Original Message-----
From: Tom Weber <pve@junkyard.4t2.com> 
Sent: Monday, September 7, 2026 12:46 PM
To: pve-user@lists.proxmox.com
Subject: Re: Migrating a 10TB VM from ESXi to PVE/Ceph

Hi,

I'd try to make your FC SAN Storage visible to your PVE Cluster - so that it can "see" the .vmdk Files.

Next define a VM with the needed Hardware Settings (CPU, RAM, Network...) but small / empty Disks on your FC SAN Storage - choose VMDK Format.

This should give you a machine description in PVE (in /etc/pve/nodes/<NODE>/qemu-server/<ID>.conf) and vm-<ID>-disk-<x>.vmdk files on the FC SAN Storage.

Now with the VM shut down (in both worlds), you could rename the production VMDK Files to the vm-<ID>-disk-<x>.vmdk names used in the <ID>.conf file and you should be able to fire up the VM on the PVE cluster working with the vmware disks on the FC SAN (current Data).

Once the VM is running you could do a live migration (move Storage) of the VMDK disk to Ceph or whatever other PVE Storage.


 From time to time I did something similar and I dont see why it shouldn't work but please test and experiment with smaller VMs first - and if possible work with Snaphots/Backups whatever before mounting the production .vmdk files in PVE.

Regards,
   Tom


Am 07.09.26 um 09:08 schrieb Fabien Sirjean:
> Dear all,
> 
> After successfully migrating most of our infrastructure (300+ VMs) from VMware ESXi to PVE, we're struggling with one last critical VM: our BlueMind/Cyrus mail server (~800 users), with a 10TB virtual disk.
> 
> Our setup is:
> 
> Source: ESXi → Dell EMC VPLEX (FC) → Dell Unity
> Target: Proxmox VE → Ceph RBD (dedicated cluster)
> 
> We tried the Proxmox ESXi importer, but I/O performance degraded so much during the live import that the mail service could not sustain its normal workload.
> We then stopped the service to complete the migration, but the import eventually failed after almost 3 days, forcing us to start over.
> 
> Application-level migration/synchronization is not an option, so we're looking for a block/storage-level approach with minimal downtime.
> 
> Since the datastore ultimately resides on our FC SAN, could we bypass ESXi entirely and temporarily present the relevant VPLEX/Unity LUN to a PVE or migration host, mount/access the VMFS datastore read-only, and copy/convert the VMDK directly to Ceph RBD?
> Even better, is there a practical way to perform the bulk copy while the VM is running, then only synchronize the changed blocks during the final cutover?
> 
> Any advice or experience with direct VMFS access, VPLEX/Unity snapshots/clones, or other approaches for this kind of migration would be greatly appreciated.
> 
> Thanks!

RE: Migrating a 10TB VM from ESXi to PVE/Ceph

[Andrei Boros]

I was thinking of the same thing for a esx->pve migration scenario we have.

Still in the concept stage until hardware is fully connected, so will be back with feedback.

On 7 September 2026 22:35:22 EEST, Adam Thompson <athompso@athompso.net> wrote:
>I went with NFS precisely because FC storage isn't generally something
>PVE can mount - it's block storage, so it'll have (most likely) a VMFS5
>filesystem and raw files with extents inside it, not VMDKs.  Pivoting
>through an NFS server flattens everything into a VMDK, which PVE can
>then mount and see and use.
>-Adam
>
>-----Original Message-----
>From: Tom Weber <pve@junkyard.4t2.com> 
>Sent: Monday, September 7, 2026 12:46 PM
>To: pve-user@lists.proxmox.com
>Subject: Re: Migrating a 10TB VM from ESXi to PVE/Ceph
>
>Hi,
>
>I'd try to make your FC SAN Storage visible to your PVE Cluster - so
>that it can "see" the .vmdk Files.
>
>Next define a VM with the needed Hardware Settings (CPU, RAM,
>Network...) but small / empty Disks on your FC SAN Storage - choose
>VMDK Format.
>
>This should give you a machine description in PVE (in
>/etc/pve/nodes/<NODE>/qemu-server/<ID>.conf) and vm-<ID>-disk-<x>.vmdk
>files on the FC SAN Storage.
>
>Now with the VM shut down (in both worlds), you could rename the
>production VMDK Files to the vm-<ID>-disk-<x>.vmdk names used in the
><ID>.conf file and you should be able to fire up the VM on the PVE
>cluster working with the vmware disks on the FC SAN (current Data).
>
>Once the VM is running you could do a live migration (move Storage) of
>the VMDK disk to Ceph or whatever other PVE Storage.
>
>
>From time to time I did something similar and I dont see why it
>shouldn't work but please test and experiment with smaller VMs first -
>and if possible work with Snaphots/Backups whatever before mounting the
>production .vmdk files in PVE.
>
>Regards,
>   Tom
>
>
>Am 07.09.26 um 09:08 schrieb Fabien Sirjean:
>> Dear all,
>> 
>> After successfully migrating most of our infrastructure (300+ VMs)
>from VMware ESXi to PVE, we're struggling with one last critical VM:
>our BlueMind/Cyrus mail server (~800 users), with a 10TB virtual disk.
>> 
>> Our setup is:
>> 
>> Source: ESXi → Dell EMC VPLEX (FC) → Dell Unity
>> Target: Proxmox VE → Ceph RBD (dedicated cluster)
>> 
>> We tried the Proxmox ESXi importer, but I/O performance degraded so
>much during the live import that the mail service could not sustain its
>normal workload.
>> We then stopped the service to complete the migration, but the import
>eventually failed after almost 3 days, forcing us to start over.
>> 
>> Application-level migration/synchronization is not an option, so
>we're looking for a block/storage-level approach with minimal downtime.
>> 
>> Since the datastore ultimately resides on our FC SAN, could we bypass
>ESXi entirely and temporarily present the relevant VPLEX/Unity LUN to a
>PVE or migration host, mount/access the VMFS datastore read-only, and
>copy/convert the VMDK directly to Ceph RBD?
>> Even better, is there a practical way to perform the bulk copy while
>the VM is running, then only synchronize the changed blocks during the
>final cutover?
>> 
>> Any advice or experience with direct VMFS access, VPLEX/Unity
>snapshots/clones, or other approaches for this kind of migration would
>be greatly appreciated.
>> 
>> Thanks!

-- 
ing. Andrei Boros
Serviciul IT&C
0745-115721