I understand you can not physically map a vproc to a physical core unless you use something like cpu affinity; however in planning I don't want to exceed that best practice given I want the end user experience to be as good as it can be. My goal is to plan X # of VMs per physical box based on vprocs matching up to cores -- 1 to 1.
The consultants at VRC have done a pretty good job testing TS/XenApp running on VMware and posted their testing scenrios with the associated results. Their tests were running on the AMD chipset though and not the latest Nehalem chipset from Intel. Supposedly Intel said that their HT technology is much better this time around. With HT enabled on a Dell R710, dual proc quad core Nehalem chips, we would have 16 logical procs. The big question I have is HT that much better this time around? Will the TS user experience degrade if I exceed that 1 to 1 vproc to physical core relationship in thinking I could go to 16 vprocs per physcal box versus 8 vprocs.
The other thing I didn't mention is that I am in a 32-bit TS/XenApp environment, so the plan is to stick with either 1 vproc or 2 vproc (most likely) VMs. The memory limit is hit before the proc is utilized.
