Re: [chrony-users] Comparing timekeeping performance with chronyd to ntpd

[Miroslav Lichvar <mlichvar@xxxxxxxxxx>]

On Fri, May 03, 2019 at 08:37:17AM -0700, Bill Unruh wrote:
> 
> Ah, I had thought it was a comparison also of ntp/chrony. You have done those
> comparisons as well I believe.

In my tests with a PPS refclock (jitter about 1 microsecond), the
RMS offset of the system clock synchronized by chronyd is up to about
10 times smaller than with ntpd using the same polling interval. That
is with a less stable clock and/or rapid changes in the temperature.
With a more stable clock the difference is smaller.

> The question is how much smaller. And do you have any idea how much of the
> jitter comes from the software reading the system clock, and how much is
> external jitter?

With a typical GPS PPS signal on serial port or GPIO, I suspect most
of the jitter is due to variable latency in handling of the interrupt.
The kernel can read the clock in just few tens of nanoseconds, but the
point when that actually happens is not stable. When reading the PPS
in a polling mode, the jitter is significantly smaller.

The same applies to the kernel receive timestamp. Variable interrupt
latency and interrupt coalescing add jitter. With a poll-mode driver
the jitter is much smaller.

-- 
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.