| Re: [chrony-users] Measuring clock offset results |
[ Thread Index |
Date Index
| More chrony.tuxfamily.org/chrony-users Archives
]
On Wed, 19 Jul 2023, Miroslav Lichvar wrote:
[CAUTION: Non-UBC Email]
On Wed, Jul 12, 2023 at 09:52:11AM -0700, Thangalin wrote:
https://ibb.co/album/1Z824p
Are there any other ways we could help chronyd stay below 20 microseconds
(e.g., optimized build) that may have been overlooked (besides using GPIO)?
If you zoom on those spikes and see that they are just single
samples, it's probably not an error of the clock, but rather a
measurement error (e.g. due to interrupt delays). You should implement
some filtering in your program, e.g. use the median value from a
number of consecutive samples (e.g. 5). That's what both ntpd and
chronyd are doing with the data they get.
Also you could eliminate one or 2 data points which are way off (say 2
standard deviations or two deviations ( Sum_i (|t_i-t_median|)/N) from the
median. Such events Mills calls popcorn events-- they are not gaussian
distributed and are probably caused by single transient events (eg like
interrupt delays, or path delays (eg the routing algorithm suddenly sends a
packet from NY to Washington via Beijing for some unknown reason)).
So getting rid of such popcorn events can improve the signals a lot.
--
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.
--
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.