[ntp:questions] Does NTP cause an abrupt/sudden change in an orphan's system time?

unruh unruh at invalid.ca
Thu Aug 9 14:59:14 UTC 2012


On 2012-08-08, Harlan Stenn <stenn at ntp.org> wrote:
> unruh writes:
>> On 2012-08-08, Arpith Nayak <arpithnayak at gmail.com> wrote:
>> > I have a server that is in the orphan mode i.e. it was not connected to the
>> > internet (and thus the various public NTP servers) when I booted it up. Now
>> > if I install ntpd oin this server and run it, will the first instance of
>> > ntpd cause a sudden change in the system time as well as timestamps or does
>> > NTP cause a slow gradual change so that it syncs up with the public server
>> > over a period of time?
>> 
>> If the time is out by less than 128msec the procedure will be
>> gradual. If it is out by more, ( but less than some 10s of seconds) it
>> will suddenly jump. If it is out by even more, ntpd will shut down
>> completely (unless you started it with the -g flag and this is the
>> first jump).  If you want more continuous behaviour, use chrony.
>
> How is chrony more continuous if it shifts between time sources that
> differ?

a it does not by default jump, and it slews the clock faster than 500PPM
if the time is out by a lot.

>
>> ntpd makes a great to-do about continuity, but jumps ifthe time is out by
>> more than 128ms at any time. That part of ntpd is an incoherent mess. 
>
> One person's signal is another person's noise.
>
> There are tradeoffs involved, I think you know this, and you are
> ignoring them.
>
>> (Note that it never slews by more than 500PPM either, unless it jumps (
>> infinite PPM). Linux at least has the ability to slew the clock by up to
>> 100000PPM (.1 sec/sec) but ntpd does not make use of it. )
>
> And there are tradeoffs here too and you know this and you are ignoring
> them.

As does ntpd.

>
> H



More information about the questions mailing list