WinPE fails to load on NFS datastore

[nickwhite44]

Hi all,

I had a strange experience today in our lab. I am experimenting with NFS datastores (on a NetApp 2050). I've loaded a bunch of RHEL 5.2 VMs and was moving to Windows 2003 Std. My company uses a WinPE-based boot image to load the Windows server OS and it wasn't working for me. The symptom was that the VM would start up, start loading Windows, load a bit of graphics as the background (gray), but then reboot before a CMD shell started. I'm not a Windows expert, so I consulted other sysadmins and they all felt like a disk driver was not loaded in the WinPE image causing the boot to fail. Normally what would happen is that a CMD shell would start up and I could run a variety of disk partioning, image restoration commands, but that CMD prompt never made it.

This WinPE image is the same that we use in our production VMware environment which runs on FC SAN. I was busy explaining that the VM didn't see any of the underlying hardware so that didn't matter that I was using an NFS datastore - if the LSILogic driver worked on the FC SAN, it should be working on the NFS Datastore. So I loaded a generic Windows 2003 Std x86 ISO and restarted the VM - it loaded fine. We let it go to the point where it formatted the C: partition and we powered the VM off - so now we have an NTFS formatted vmdk. Then we swapped ISOs back to the WinPE image, loaded the formatted vmdk and it booted fine - no problems.

We started talking about whether or not the vmdk was pre-allocated or not and I had to admit that it was a thin-disk on NFS. Of course the other sysadmin, who are really opposed to this NFS idea, said that's the problem. So, the next experiment was to try to load a new VM with the WinPE ISO using a local datastore (local disk to the ESX server) - It worked with no problems. So WinPE fails on an unformatted NFS vmdk, but works on a VMDK with a pre-existing filesystem.

So, here I am wanting very much to continue with my NFS proof of concept and I have a hurdle. What is the problem? Is it obvious to you guys?

Thanks in advance,

Nick

[Rumple]

well, I can't offer any advice on what you are seeing, but you can pre-allocate the disk in thick format using the command line if you think thats the problem. Personally I use vm templates so once I build one its 5 minutes and i have another VM running using Vc cloning.

As for nfs vs FC I can tell you that I built an entire farm with 50+ vm's running on 5 different lun's on my CX700 Clariion SAN and with a merger we started moving to NFS mounted lun's and can tell you I will never go back to FC unless forced.

Herei s why I love NFS now...even though I was the EMC SAN admin...

The ease of mounting a LUN without worrying about zoning

The ease with expanding a volume on the fly

The fact that netapp supports dedupe, which in an environment with many systems the same ends up with upwards of 60% disk savings

The fact that there is no SCSI reservation problem because the locks are per file, not LUN based

The fact that instead of trying to build to 10-15VM's per LUN I can run hundreds of VM's on a lun and not worry about sizing as much (other then spindle)

Did I mention no zoning....

Utilizing Snapmanager for vmware (which is dirt cheap by the way) I can snap my entire ESX farm over to DR without worrying about it (with obvious file level precautions)

I will be dragged kicking and screaming back into the FC or ISCSI world now (with the possible exception of a FC/iscsi lun for VM page files)

[scerazy]

Rumple, which NFS appliance do you use?

Seb

[Rumple]

we use netapp 3050's....