Hi,
Thanks for the pointer. The article is very enlighthening. I speed read it and plan to slow-digest it next couple of days
I notice that I/O are very expensive on VT (current implementation in the least). However, given that most of the modern peripherals are capable of memory I/O, especially those disk controllers (e.g SCSI), is it not true that as long as any virtual device driver keep to using memory I/O once the initial set up is done, we can avoid a lot of the penalty from I/O on a VT based virtualization ? The point here is that the benchmark is enlighthening, but in practical usage, will any those penalty be suffered ?
One thing for sure, the close to native graphic intensive performance does explain why I observe that running windows on VT enabled do feel faster. On a desktop, it matters, but for servers, I agree it does not matter.
I have this suggeston : Why don't VMware VMM do some form of instrumentation on-the-fly, keep simple counters on those expensive VT instructions, allow the VMM to switch from VT enabled to pure software virtualization.
Thanks.
p/s:
Note that I did not failed to run 64-bit guest, but was trying to know how to verify that the guest is in VT mode (beside spendign a lot of time doing benchmarks to tell).
Message was edited by:
kawika