Re: [chrony-users] Accurately measuring clock drift

[ Thread Index | Date Index | More Archives ]

On Tue, Jun 20, 2023 at 10:16:36AM +0200, Rob Janssen wrote:
> That serial port example uses gpsd.  I wonder how that compares to using
> "ldattach 18 /dev/ttyS0" and using /dev/pps0 in chrony.
> That avoids the path through a user process and via SHM (which has only
> microsecond resolution).

SHM has a nanosecond extension that is supported by both chronyd and
gpsd. The example uses SOCK, which may work slightly better than
SHM and PPS as it avoids the driver polling (the sample is available
to chronyd very quickly after the PPS interrupt). But this difference,
if there is one, would be much smaller than 30 microseconds.

> I use chrony this way on some servers (also for simulcast) and I do poll and
> graph the offset of System Time displayed by chronyc tracking.  These graphs
> show that the offset peaks differ quite a lot by type of machine, with e.g.
> some HP ProLiant machines peaking up to 5us offset while (older and slower)
> Dell servers peak up to 200ns.  No idea why.

It could be due to different power-saving mechanisms. For best
stability the CPU needs to be set to a constant frequency and the
power saving states need to be disabled (e.g. idle=poll).

Miroslav Lichvar

To unsubscribe email chrony-users-request@xxxxxxxxxxxxxxxxxxxx 
with "unsubscribe" in the subject.
For help email chrony-users-request@xxxxxxxxxxxxxxxxxxxx 
with "help" in the subject.
Trouble?  Email listmaster@xxxxxxxxxxxxxxxxxxxx.

Mail converted by MHonArc 2.6.19+