NPIV & Clustering

[AndyMcM]

Hi All, I am working on a project at the moment that I am trying to impliment backend file clustering with GFS.

Am hoping that NPIV might be able to help out with a problem. Has anyone tried to attach more than one VM using NPIV to a single LUN?

What I am hpoing is that I can create a new Storage Group and Zone in the VMs to gain access to a group of LUNS and then use GFS to share data between multiple servers.

Can anyone help out with this?

A.

[mreferre]

Andy,

I should say upfront that I am still warming up with 3.5 but, if I remember well, NPIV is not supported with RDM's ..... and RDM's are typically a sine-qua-non for clustering setups.

So you either might have support troubles with (using VMFS for clustering scenarios) or you loose NPIV support (using RDM's).

If I can ..... what is the reason for which you want to use NPIV? Why don't you just map the LUN's to the ESX host and you assign the LUN's (as RDM's) to the selected vm's which comprise the cluster nodes? Just out of curiosity....

Massimo.

[AndyMcM]

Hi Massimo,

I am trying to figure out how best we can get three plus RHEL5 boxes accessing the same data a GFS partition.

Was hoping that using NPIV I would be able to Zone in three plus VMs together and then give them access to the same LUNs.

I have used RDM to do map VMs directly to LUNs but ESX 3.02 has a limitation that you cannot do more than two maps to the same LUNs.

Have also investigated making the SCSI sharable and then accessing the devices that way too but not sure how reliable that is.

Any thought would be great thought,

Cheers.

A.

[mreferre]

Andy,

I am not a fan of NPIV .... and I can't claim to be a fan of clustering on VI3 either ....

Having this said ... the best option, in my opinion, to do what you are trying to achieve is to use RDM's w/o NPIV (due to those limitations I was referring to ... I should still check if it's true).

I didn't even know there is this limitation of having no more than 2 vm's pointing to the same LUN...... is it a supportability limitation or a functional limitation ?

Massimo.

[StephenTJ]

The NPIV functionality is only available to guest servers with RDM disks. A virtual machine with NPIV configured just allows for that virtual machine to route all its traffic to its RDM disks through its virtual port on the host's HBA. This allows for the SAN Admin to control access to a LUN on a per virtual machine bases, but the documentation says any LUN you give to a virtual machine has to also be assigned to the ESX host. I was curious if you could assign the LUN to the virtual guest on the SAN side and see if the guest saw it without using a RDM, unfortunately I cannot do any playing with NPIV as my hosts think my fabric doesn't support NPIV (probably because of HP Virtual Connect, not sure just yet).

[mreferre]

I apologize for the confusion.

True ... it's the other way around ... NPIV only available through RDM's not VMFS. Sorry for the confusion ...

Massimo.

[AndyMcM]

Cheers guys for the response, looks like it may be a no go. Going to get 3.5 installed and our FC Switches upgraded to use NPIV and see what I can do.

Thanks Again.

A.

[brianplume]

Stephen,

I, too, have run into the same issue you are having, and came to the assumption that it is Virtual Connect as well. My guess is that since Virtual Connect is already doing NPIV, adding another layer of NPIV at the virtual machine level is too much for it. Just wondering if you ever got clarification that this is the issue?

Take care, and good luck.

[StephenTJ]

After going through 4 different HP techs trying to explain what I though was the issue I did finally get clarification from an HP tech about this. I told them that I thought the issue was that the 16 downlink connections did not support NPIV due to the fact that they are a 1:1 connection ratio (one server HBA per port). The tech I spoke with confirmed what I thought, the 16 downlink ports do not support NPIV, only the 4 uplink ports do. Hopefully at some point maybe a firmware update will allow for this to be supported but until then we are out of luck I think.

If I get any further information about this I will post it.