[ntp:questions] Seeing large jitter and delay in recent ntpd and recent Solaris
hart at ntp.org
Thu Nov 10 01:49:16 UTC 2011
On Wed, Nov 9, 2011 at 23:44, Chuck Swiger <cswiger at mac.com> wrote:
> On Nov 9, 2011, at 2:47 PM, Chris Albertson wrote:
>> That will do little good and maybe some harm. ntpd reads time stamped
>> input. Even if this sits in a buffer unprocessed it's OK because the
>> critical work, the stamping is done inside an interrupt handler.
> Agreed, Solaris ought to support and be using SO_TIMESTAMP, so it
> really doesn't matter whether ntpd itself is running at elevated priority.
Agreed assuming the source of time is another NTP server.
>> If you are concerned about offset the solution is to use a good local
>> reference clock
Not to be confused with the undisciplined LOCAL reference clock ;)
However, note SO_TIMESTAMP doesn't help with a non-PPS refclock, in
that case elevated priority can help, but I still would never run ntpd
at higher priority than the OS interrupt-processing thread(s)
(applicable only if the interrupt processing is in fact handled by a
scheduled thread, which Chuck seems to suggest is the case for
> For best timekeeping, also setup PPS discipline.
With a PPSAPI refclock, the situation is much like with SO_TIMESTAMP
-- there's not much, if any, benefit to running ntpd at elevated
priority as the PPS timestamp is taken in the interrupt handler.
More information about the questions