On Fri, Feb 06, 2004 at 07:00:53AM -0600, Rich Smrcina wrote:
> That varies based on processor.  The bigger the processor the smaller
> load the timer code puts on it.  On a G5 I saw about .3% CPU Utilization
> from an idle machine without the patch.  On a z800 it drops to about
> .1%.  On a z900 it is probably negligible.  But 100 machines on a z900
> without the timer patch would be measurable.

Also remember that the real gain is for the rest of the system, because
a guest who's popping his timer never really gets idle.  For a
production workload where most machines are busy most of the time,
that's probably OK.  On the other hand, for a development system (such
as ours) where there are a bunnch of virtual machines with various
projects on them, only a couple of which are active at any time, I would
rather have those guys really really idle if I'm not using them.

On an H70, I can confirm that we see about .3% CPU per guest when it's
otherwise idle, if it's doing a  100Hz timer pop.

Adam

Reply via email to