We are currently running a Hyper-V 2012 R2, 2 Node Cluster which contains 40 virtual machines across 7 CSV volumes.
I was looking at improving performance and noticed that we did not have the CSV cache setting set on the cluster. The value was set to 0 and CSV cache enabled on all CSV volumes as per Microsoft default settings.
I then decided to increase the value of the cache from 0 to 1GB as per Microsoft recommendations:
(Get-Cluster). BlockCacheSize = 1024
After I enabled it I saw something very strange in that I was unable to remote desktop on some of the servers in the cluster. It looks like they were the server that owned the CSV and VM therefore having the cache working for that virtual machine. The VM was up and working for user connectivity, however as an admin when ever I tried to access it I got a black screen.
I also noticed that when I tried to log on using the console session in the failover cluster manager it didnt display the logon Gina after a CTRL+ALT+DEL.
At this point I was concerned about the stbaility of the system so I changed the CSV cache back to 0 again. Then everything came back to life and I was able to RD and get a console session again on the servers.
Has anyone ever seen this before? All the blogs I was reading suggest there is no issues enabling this and only performance increases. I didnt restart any nodes after this was changed because Microsoft articles state the memory cache allocation in Server 2012 R2 is dynamic and does not require a node reboot.
Our storage backend is a SAN and all the CSV volumes are connected using multipath and iSCSI.