Hi all,
We plan to build hundreds of VM using VMFS for OS and binaries and physical RDM for data.
Some VM will need until 14 Luns for data, that's why we plan to use RDM instead of datastores.
The San administrators would like to isolate the RDM stream to facilitate performance debuging. So using NPIV could be a good way.
As far as I understood to create theses VM,
1. I will use a boot disk stored on VMFS DS and a RDM disk.
2. I will configure 2 WWNN and 2 WWPN on the VM using Virtual Center.
3. I will give the WWNs to the SAN administrators.
4. Once the zoning and lun declarations are done on the SAN, I can start the VM.
If ESX can access the Luns using the WWNs, it will use it. If not, it will use the WWNs of its physical hbas.
Now, a scenario :
Two VM are requested with a RDM lun each.
The SAN team provide me 2 datastores
I create 2 VM vmA and vmB with 2 RDM : rdmA and rdmB
I give the WWNs and the luns reference to the SAN administrators.
If the SAN administrators link the WWNs of vmA to lun rdmB and WWNs of vmB to lun rdmA, I will not notice an error when I will start vmA and vmB because both VM will use physical WWNs of the ESX.
The only way to check is to verify using : esxcfg-scsidevs -a.to check if the 2 WWNs of the VM are listed.
So it's not very satisfying for a big production environment.
Is it correct ?
Can we remove the WWNs of the physical HBA of the ESX once the RDM disks are zoned ? In fact, the ESX do not need to access the RDM Lun any more.
Perhaps, it could be interresting to allocate WWNs to the VM, then using the WWNs of the VM to add RDM disks. Perhaps in vSphere 5 ?
Thanks for your advices
