I wanted to get this information out here, as we’ve had a few customers run into this. The issue does not appear to affect core cluster functionality, but does affect the appearance of the cluster.
First, one of my colleagues, Tim McMichael, has a blog that details the different ExBPA rules that can be ignored for Windows 2008 clusters (both Exchange 2007 and Exchange 2010). That blog is located here:
I’m not sure if the rules ever got updated for Exchange 2007, but they definitely were not updated for Exchange 2010 RTM (Oops!). These rules will be fixed with Exchange 2010 SP1.
Now, on to the issue.
If you run ExBPA on Exchange 2010 RTM, one of the issues detected (under All Issues) is that the Quorum Log size is too small (see below screenshot from my lab)
You’ll note that the recommendation is to change the current setting from 1024 to 4096.
http://technet.microsoft.com/en-us/library/aa995830(EXCHG.80).aspx details the recommended change and where to modify this setting. If you are astute, you’ll note that the text only discusses Windows 2000 and Windows 2003. That’s because this setting no longer applies to Windows 2008. Unfortunately, you won’t find anything currently documented about this (I’ve asked our friends over in Windows Clustering to please get something documented).
You’ll also see the same recommendation if you run Test-SystemHealth (the Shell version of ExBPA)
So, let’s say you are running Exchange 2010 in a Database Availability Group (DAG), and you decide to make the change. Remember that for a DAG, you have to specify a Witness server during creation. This results in the Quorum model being set to Node and File Share Majority, which will show as follows:
As soon as you modify the MaxQuorumLogSize registry value to anything but the default of 1024, and refresh the Failover Cluster Management interface, you will notice that the Quorum model changes to Other.
As stated previously, this doesn’t affect the actual functionality of the cluster, but if I was a customer, and encountered this situation, I sure wouldn’t be happy, especially seeing that a Quorum model of “Other” isn’t documented anywhere, and isn’t even shown as a valid Quorum model.
The resolution to this is easy. Change the MaxQuorumLogSize back to 1024. No restarts required – just refresh the Failover Cluster Management interface. The Quorum model should change back to “Node and File Share Majority (\\UNC of FSW)” as expected.