[ntp:questions] like a kid with a new toy (PPS jitter)

Dave Hart davehart at gmail.com
Sat Feb 7 20:07:27 UTC 2009


On Feb 7, 6:51 pm, ma... at ntp.isc.org (Danny Mayer) wrote:
> Dave Hart wrote:
> > It's running with a patch to lock the main and timer threads to the
> > 2nd CPU using SetThreadAffinity.  I haven't tried locking the async I/
> > O thread as well, but I will now.  It's also running with a patch to
> > the interpolation code which is more selective about baseline counter/
> > time pairs it uses.  There is a faster server on the same LAN I could
> > try as well, but I figured slower is better for understanding how bad
> > the user-mode PPS implementation is.
>
> You only set the thread affinity if it really needs it otherwise it
> doesn't really matter which CPU it's using. Certainly the I/O thread
> should not be set.
>
> Danny

I'm afraid the I/O thread does need to be nailed down because it, like
the main thread, uses get_systime() which uses gettimeofday() which
calls QueryPerformanceCounter and relies on consistent results as
compared to the timing thread which sets the baseline time/counter
correlation.  This is what I just said in https://support.ntp.org/bugs/show_bug.cgi?id=1124#c10

Cheers,
Dave Hart




More information about the questions mailing list