DragonflyBSD kernel clock improvements
Don Bowman
don at sandvine.com
Sat Jan 24 10:55:01 PST 2004
> From: Bill Moran [mailto:wmoran at potentialtech.com]
>
> Hello all,
>
> I saw this recently:
> http://www.dragonflybsd.org/Docs/nanosleep/
>
> I was wondering if anyone on the FreeBSD team has looked at
> this. It doesn't
> appear as if any recent change have been made in the FreeBSD
> tree regarding
> this.
>
> It's probably not a big deal, but I just thought I'd point it out.
I think its important, we do have some persistent time problems
(http://lists.freebsd.org/pipermail/freebsd-stable/2003-November/005194.html
)
which seem to require kern.timecounter.method=1 to work around.
I think an increased focus on clock accuracy and jitter can't
be a bad thing, even if FreeBSD is not a real time system.
increasing the repeatability is good for DEVICE_POLLING, for
tcp timers, etc. As more and more things move out from under the
giant lock, events can become increasingly deterministic.
Glad to see someone is putting a scientific method to the
problem, it provides a good baseline for increasing the accuracy
of measurements.
--don
More information about the freebsd-current
mailing list