PVSCSI and Windows 2008 R2...

-
Bleeding edge?  So you will proboably know what PVSCSI is all about

We had two days of worry - our Exchange 2010 Mailbo server was showing events of having a corrupt database.  The event description pointed at the storage. 

ESE 482:


Thus ensued a witch hunt for the actual culprit.  Storage logs reported all fine...the issue came and went.  Unfortunately when it came the databases failed over to the other node of the DAG.

We found a page on VMWare's site detailing that a high IO server could have issues when using PVSCSI and that updates to the driver are required.  The driver updates pre-requiste was of course a patch to ESX 4 too.  Not something we can do on a whim, change management anyone?

See here:
http://longwhiteclouds.com/2012/03/14/win-2k8-with-pvscsi-critical-issue/#more-801

and here:
http://kb.vmware.com/selfservice/microsites/search.do?language=en_US&cmd=displayKC&externalId=2009144

So we had to remove the PVSCSI devices.  This is actually much simpler than it would be if these were real devices and we were able to move the Database and Log LUNs to the SAS drvier without hitch we then moved the boot disk. 

As we were able to fail over the databases whilst doing this there was NO USER DOWNTIME.  Gotta love Exchange 2010 sometimes.

Did anyone notice the "high IO" in bold earlier?  This still requirement is confusing, on one hand Microsoft says the Exchange 2010 is 70% less IO then 2007 (which was 70% less IO then 2003) so in my books in not a high IO system - this is confirmed by our NetApp storage which indicates that Exchange barely rumples the IO.  On the other hand we definately saw this issue and changing the PVSCSI device to SAS has made the issue go away, which would imply that Exchange was doing high IO.  Or something on the mailbox server was doing high IO...backups?  Forefront?

The case continues...

Comments

Post a Comment

Popular posts from this blog

PXE booting, MDT and 802.1x

-
Image
Oh dear. We implemented site wide 802.1x when we moved. This has caused numerous issues which we are still tracking down and killing with fire. Again "only-when-we-needed-it" struck as we had a new starter and need to build a machine. Ah.   No PXE.  However thankfully we were able to make a bootable USB using 802.1x and get our build working again from MDT in our 802.1 environment. Here's how we did it.   Caveats:  I'm assuming you know MDT and how PXE works. We use Computer certificates as the 8021x auth method. 802.1x is working in our environment GPO is used to start the wired 802.1x service on built machines. GPO is used to configure the network profile to use 802.1x What you need Get a USB stick to take your WinPE 4GB may be enough Make sure you are running an up to date working MDT installation! Create a machine certificate which has a long expiry date, you don't want to be making USB WinPE images every month!! Export the machine
Read more

Intune installation requires a wire...or does it?

-
If you are using Intune (great!) it assumes you have a wired connection (not great) - this is more of an issue as some machine now no longer have built in ethernet ports. It is possible to connect to your wireless - albeit in a slightly awkward way.....here's how During the OOBE hit SHIFT+F10 to get a command window and type either; start ms-settings: To get the settings window or start ms-availablenetworks: To see the network fly outs. Thanks to  Using a wireless network connection with Windows Autopilot white glove – Out of Office Hours (oofhours.com)
Read more

Security Policy 1001

-
We had an annoying warning showing every 5 minutes on our Read Only Domain Controllers.  This didn't affect service but meant we were losing our logs of important events (should the have occurred) "Security policy cannot be propagated. Security Configuration Server (in services.exe) is not ready. This is probably in system reboot. Policy will be tried again in the next propagation." Source: SceCli Event: ID 101 We tracked this down to a GPO which was populating the local Administrators group. Obviously a Domain Controller has no local users or groups so this was causing the issue. Ensuring that the GPO did not apply to the Domain Controllers - yes the GPO was linked at the root! fixed the issue. It is strange that we did not see the warnings on our main Domain Controllers only on the Read Only DC's as this would have affected all DCs the same (I would have thought)
Read more