Accuracy/reliability of "top" CPU usage numbers

[Denys Vlasenko]

On Fri, Nov 14, 2014 at 5:15 PM, Cathey, Jim <jcathey at ciena.com> wrote:
> Things like 'top' have _always_ had issues with processes
> that are clock-synchronized.  Very few OS's (and I don't
> think Linux is one of them) actually account CPU time
> by use of things like the PPC TBR register (for example).
> Most do it via statistical sampling in a clock interrupt.

Yes, this is most likely the cause.

The HOHZ come which has landed in recent kernel
and which is being currently refined mostly switches
to accurate (as opposed to statistical) time measurements.

So, this problem is likely to be resolved in more recent kernels.