i have been evaluating 2012 Enterprise. we current run 2008R2 Enterprise.
there is an 18 percent degradation from a key job that we run in 2012 compared to 2008R2 64.
Here's the test environment:
1 Xeon 2650 socket 2011, 32 gigs of memory. C drive is 15k segate SAS cheetah. Second drive is OCZ Vertex 4 500 gig. The cheetah is attached to a low end adaptec controller, the SSD is attached to a LSI chip on the mother board.
Windows server 2012 GUI Datacenter was installed, then Hypervisor. Three virtual machines were created under the hypervisor. All three use the exact some sized disk in the exact same location. The target dbs, both data and log are on the SSD.
Two VMs had Windows 2008R2 64 bit with latest sps. One of those had windows 2008r2 64 bit SQL Server installed, 10.50.4000 one win2k8 server had SQL server 2012 Enterprise intalled, 11.0.2100.
The third VM had Windows Server 2012 GUI standard installed. SQL Server 2012 Enterprise 11.0.2100 was installed on that machine.
The same database was restored to each VM. The compatability level of each db was set to the most recent available one in that version of SQL Server.
Each table that partcipated in the test had all of its Indexes rebuilt and then the statistics were updated with full scan.
the the job was run.
The job reads target records out of a table containing 26MM rows. approximately 6.6MM rows are used to build a secondary table. The first part of the job builds empty records for all subsequent work. Then each record from the source table is used to update a field in one of the existing rows in the destination table.
the recovery mode of the db is simple.
Every 100 updates, a transaction is committed and another is started.
WAIT states are always writelog for all test environments.
The defaults were taken in all SQL Server installations.
it takes approximately 18 percent more time to run this job on SS 2012, regardless of the OS. All VMs are configured with the exact same resources.
It appears that at least for write intensive (OLTP) type activty that SS2012 is less effecient.
I continue to evaluate and look for possible tweaks but i was a bit surpised by this result.
I may do a bare metal (non VM) install of win2k8r2 to see if the performance difference is maintained in a non VM environment. To do that i need some more hardware though to duplicate my current test environment.
The test is part of a real initialization that is run in our production environment. The test was used to shoot down VMWARE hosting of key SQL databases. The folks at VMWARE could never approach the speed of some of our faster non production (DEV/QA) servers.
One anecdotal aside, the Hyper-V performance at the disk level appears to be near as good as a bare metal OS solution.