Guess what? Very little has been documented so far.
The title comes out of the TechNet Hyper-V forum and has finally been dealt with on Tony Voellm's MSFT blog. However, it isn't the entire picture - there is more going on here than a simple NUMA bug.
What I have been able to uncover is this:
- The parent partition reserves 256 MB RAM for itself that can never be consumed by a VM.
- Each VM that is running carves a physical slice of RAM out of the hardware.
- Each VM causes an additional 32 + 8 Mb of RAM to be used by the paraent partition for over head and management (not unique to Hyper-V).
Now, all that being said, this is what I have been able to uncover and verify in some way.
Server sizing recommendations follow that of Server 2008, plan on at least 512 Mb of RAM for the parent and then add on for the children (the actual VM RAM plus overhead).
Now - to the issue.
The NUMA bug:
Tony Voellm talks about a NUMA bug here:
This article keeps pointing to caching..
There is also a few threads in the TechNet forum that have resolved this issue in another way.
the root thread is here:
The solution on this thread? - disable automatic pagefile management in the parent partition.
Again, we go back to paging (or caching..)
So, something is going on under the hood that I am sure will be fixed but for now think ahead and take a look at the articles.
Personally, I would try changing the page file setting or relying on Remote Management before applying a patch, but that is just me.. :-)