Anyone done MSCS with MS iSCSI initiator inside ESX VMs?

L

lopoetve

Extremely [H]
Joined
Oct 11, 2001
Messages
33,938
I'm tinkering with something here at work - since MSCS on iSCSI RDMs isn't technically supported, I'm curious if I can get it stable with the MS initiator inside the VM (protip: MS initiator is actually faster than the RDM passthrough... by a VERY significant amount). Problem is, I keep losing connection to the luns it seems... or corrupting them >_<

I'm just trying to do single path at the moment, no MPIO, and I haven't even put anything on the cluster. Farthest I got was both nodes up, but then it crashed and would never start again.

Just curious if anyone else has tried this before I keep reinventing the wheel. Of course, this is all within workstation, so it may not work at all...

edit: Oh, and ESXx2 in WS for the host... Win2k3 R2 Enterprise VMs. Openfiler SAN. Just curious if anyone's had it work at all, so I know if this is feasible IRW before I spend any more time on it.
edit2: Yes, I know this is crazy.
 
Sounds like a test I did 6 months ago, but I did it minus MSCS, and that's a signifcant portion of your test config. Here was mine:

Physical box (Vista x64 Ultimate) running WS 6.5
2x ESX 3.5 VMs under WS6.5
1x OpenFiler iSCSI target VM under WS 6.5
1x VC Server VM under WS 6.5

ESX hosts were both mapped to the OpenFiler VM for VMFS storage, hosts being managed by single nested VCServer VM

OpenFiler VM had a 80GB "LUN" given to it (virtual disk in WS6.5)
VCServer>Datacenter containing ESX 3.5 VM Hosts>4x Windows XP double nested clients

Took forever to setup. Double nested VMs are pretty nasty. Sounds like your just swapping out the XP double nested VMs for MSCS double nested VMs, right? Should work, but won't be easy. The CS portion is gonna make it harder.
 
Pretty much, but the MSCS vms are getting the Quorum and shared disk via the MS iSCSI initiator mapped through a second/third network card to my dual-pathed iSCSI Target (same network cards as the vmkernel ports for the iSCSI traffic, and yes, I'm multipathing with some custom scripts :)). That's where the issue seems to be (namely because openfiler doesn't support physical mode RDMs). I keep eating the storage devices - they get corrupted.


Oh, and all of this is running on a laptop, so it's slow as SNOT.
 
what if you remove the MPIO scripts? I ran my config on my laptop too, but I could venture A guess that my laptop is probably got more under the hood than the one you're trying this on...although mine would probably go to it's knees too.
 
Figured it out - it was the cluster header. I screwed it up once and wiped the vms and recloned them from my template, but I kept forgetting that the cluster installs a header on the disk so only things in the cluster can see them. This is what kept happening. I had to ignore the warnings on the second node, push through to the final setup, and then it'd work.

I did it! Works quite well too :)

MPIO made it duplicate disks like mad, pulled those too :)
 
you're referring to the cluster header on the quorum drive correct? Where did you have the MPIO scripts? I would assume in the vmx files for the ESX single nested VMs, correct?
 
sabregen said:
you're referring to the cluster header on the quorum drive correct? Where did you have the MPIO scripts? I would assume in the vmx files for the ESX single nested VMs, correct?
Click to expand...

quorum and shared drive - the goal of the cluster is to be a disk share, so it tagged the shared storage drive as well.

The mpio goo was just whatever MS installed by default. They don't have anything that goes in the VMx file that I know of.
 
You must log in or register to reply here.
Back
Top