[There's a reason that Yoda is the unofficial mascot of SBS. Size indeed matters not.] Dynamic Memory and SP1 - THE OFFICIAL BLOG OF THE SBS DIVA
Wed, Feb 23 2011 0:16 bradley

Dynamic Memory and SP1

So the jury is still out on dynamic memory and SBS 2011.

dynamic memory and Exchange or SQL:
http://social.technet.microsoft.com/Forums/en-US/winserverhyperv/thread/97689834-5657-4c87-b520-31c994bf825f

"Great question. We're working with these teams to have guidelines on configuring Dynamic Memory for these workloads.

SQL Server announced full support for Dynamic Memory (http://support.microsoft.com/kb/956893). We'll start having blog posts about the initial guidelines in the blog posts pointed by above link and eventually publish a whitepaper around it. Important thing is to set Memory Buffer = 5% for SQL workloads.

Exchange Server is a little bit different. Mailbox Role in Exchange Server does not change its memory allocations on the fly. So Dynamic Memory will not provide many benefits for this role. Other roles of Exchange can run with Dynamic Memory without any special configuration.

Thanks,

Serdar Sutay [MSFT]
Program Manager, Hyper-V "

Bottom line in our unscientific testing... that SBS 2011 doesn't budge.

Filed under:

# Question....

Wednesday, February 23, 2011 8:48 PM by Joe Raby

Wouldn't it be better if disk caching functions would be moved out of the VM to the hypervisor?  Could VM's share a common disk cache?  A lot of VM's would be caching the same data if the operating systems were the same, so couldn't some reads pull shared data from precached sectors off the physical hard disk, regardless of which VM it was from (if the data was all on the same physical disk, the cache on the drive or controller might already do part of this), and/or consolidate memory usage between VM's?

This would boil down to almost being compressed memory pages.  Does that make any sense?

When I look at Hyper-V, it doesn't allocate much to the disk cache, regardless of how many VM's are running.  However, each VM will take so much from their own memory allotment for their own disk cache.  Why couldn't this be shared?  It probably wouldn't have to be as much memory as every VM's own cache added together, I'd imagine.  However, would a VM really need it's own internal disk cache at that point?

I just think about how much similar data would be between each VM and wonder why disk access isn't treated like new multi-core CPU's - shared cache.