Re: xen on i7-12700: works with cet-no-ibt !!

[Greg Troxel <gdt%lexort.com@localhost>]

Brad Spencer <brad%anduin.eldar.org@localhost> writes:

> Greg Troxel <gdt%lexort.com@localhost> writes:
>
> [snip]
>
>> It looks like the dom0-pv vs bare metal slowdown is still huge, 70.1s
>> for PV vs 27.6s for bare metal.
>
> This may be expected to some extent.  A hit pretty much has to happen
> and it may be mostly a matter of whether or not it can be tolerated.  I
> try to run very little on my DOM0 guests, if at all possible.

Sure, but running at 90% is one thing compared to 39%.

I had not previously realized how slow PV systems were.

I'm running benchmarks on a computer with VMX/VT-d/EPT/HAP and which
thus can run PVH and PVHVM guests, and will post that.  I don't want to
conclude this seriously, but I am starting to think one can bin into old
and new, and that
 
 PV is 30% ish on very old hardware
 PV is 40% ish on newish hardware

 PVH is 66% on new hardware

 HVM is atrocious on very old hardware
 (nonpv)HVM is 65% on modern hardware
 PVHVM is 80% on modern hardware


and that therefore, one should

  test PVH/PVHVM to see if it works and if it is reasonable speed or terrible

  run PVHVM if possible
    benchmark PVH and PVHVM to be sure

  run a dom0 PVH instead of PV if possible, but take that with a huge
  grain of salt because I've never successfully booted let alone
  benchmarked it

  run your own benchmarks with your own workload on your own hardware!

  figure out why PVH is slower than PVHVM, and have others test to see
  if that's a general result, just me, or a calulation error on my part.