Win2k3 Dynamic disks + ESX virtual hosts = converts to basic = data loss

Question

ESX 3.0.1 Windows 2003 SR1 running virtually Clariion CX700 flare 19 We present the ESX servers with 512Gb luns. They present them to the virtual hosts as raw devices. They stitch them together using dynamic disk. This worked fine for a few months but 4 times in the last 3 weeks these hosts have, in the middle of the day, decided that the dynamic disks are now foreign basic disks resulting in massive data loss. Microsoft is saying it's the SAN due to disk read errors in the Windows logs (highly unlikely with 2 separate FC fabrics and 4 paths to each device we would have to have lost both fabrics at exactly the same time and for long enough for Windows to give up on the writes) Has anyone else seen this? Any thoughts?

Kiran3 · Answer

make sure that you have correct and latest driver AND firmware on the HBAs.
intermittent disconnections can mosly be due to this.

EMC is likely to advise against dynamic disks...

has the ESX host logs been looked into? was there a performance issue when the disk errors occured inside windows?

Allen Ward · Answer

A question for Kiran here though...Why is EMC likely to advise against Dynamic Disks? This is something I've never heard from them. We don't use them inside client hosts on ESX, but we use them extensively on native Win2K and Win2K3 hosts.

an_hidden_KB · Answer

decided that the dynamic disks are now foreign basic
disks resulting in massive data loss.

i'm not an expert on MS matters but this does not sound right. best to my knowledge only dynamic disks can be foreign. converting to basic results in data loss sure, but i can't see how this can happen without an administrative decision.
anyway, if the use of dynamic disks is not really mandatory I'd stay with basic disks on SAN. also performance-wise, it should be far more efficient to use metaluns on the clariion.

Allen Ward · Answer

I'd tend to agree with Ain. I'm not sure what is causing the problem, but one good solution would be to put the pieces together in a MetaLUN before presenting it to the host. This way the host will not have to deal with Dynamic disks and they still get the advantages of the striping (or concatenation) that they want.

Kiran3 · Answer

Even I am not sure about the root reason...but EMC likes to see basic disks on Windows hosts and MetaLUNs on array. Basic disks can still be expanded using diskpart

an_hidden_KB · Answer

not sure if this helps but EMC recommends adding (if not present) a TimeOutValue DWORD key to the hive HKEY_LOCAL_MACHINE\System\CurrentControlSet\Services\Disk\ in registry and set it to 60 seconds (3c in hex).

Allen Ward · Answer

Can you point to any documentation to support this. I'm not disagreeing, I just want to know if they are going to pull out some obscure document some day and say " Well, too bad about your data, but you should have followed this rule".

I don't remember ever seeing anything about this in the CLARiiON Best Practices documents.

TyfoidKid · Answer

They were using Dynamic disk because we had presented the ESX cluster a group of 512Gb luns and the Windows server admin wanted one device so they stitched them together with dynamic disk to make a 1Tb lun. We didn't want to give them 1Tb luns meta or otherwise because if you start having a performance issue with a lun that big it takes a week to migrate it to some other part of the clariion (hey EMC where's my in frame virtualization so I don't have to keep track of all this??)

TyfoidKid · Answer

ESX has been looking into the logs. Don't know that they found anything. No issues on the Clariion or the FC fabric when this event occurred.

CLARiiON

Was this post helpful?