[ntp:questions] NTP not syncing

unruh unruh at invalid.ca
Sun Dec 8 05:42:39 UTC 2013


On 2013-12-08, Harlan Stenn <stenn at ntp.org> wrote:
> unruh writes:
>> On 2013-12-07, Harlan Stenn <stenn at ntp.org> wrote:
>> > unruh writes:
>> >> As I said, get chrony if you want much faster convergence, if you want
>> >> better control of the time offset from UTC, and if yourun Linux or
>> >> BSD.
>> >
>> > And this begs the question of "faster convergence towards *what*?"
>> 
>> UTC :-)
>
> At least you added the smiley.
>
>> > NTP looks a a number of time sources, finds a majority clique that
>> > exhibit "good" (consistent with respect to each other, as observed by
>> > the local system) behavior, and steers the clock towards that time.
>> 
>> So does chrony, except that it takes one of those (the one with the
>> smallest spread I believe) as the representative.
>
> And this is the crux of the difference in approach, as it "shifts faith"
> from the current machine to the machine being followed.
>
> Sometimes this will be right and sometimes it will be wrong.

ntp shifts faith to the machines being followed. Just as ntpd chrony
also looks at as many servers as you have configured, and tries to
figure out the best range just as ntpd does. Because chrony keeps
information about the past, it can for example decide which of the
servers has the smallest standard deviation lying within the group of
good servers. 

>
>> > As I understand it, chrony steers the local clock towards the master
>> > time source it is following.
>> 
>> And it figures out which source it is following by looking at all of
>> the sources it is following.
>
> Is there something other than the code where I can learn more about
> this?

Not as far as I know. And it has been a while since I have looked at the
code, so my understanding might be problematic. 


>
>> > Assuming this is all true (or true enough), there are times and
>> > situations where one of these approaches is better than the other, and
>> > other times and situations where the converse is true.
>> 
>> Yes, but tests that both I and Lichvar have run show that chrony follows
>> to much higher accuracy than does ntpd.
>
> And again, it is not clear that chrony is following the better source of
> time.

Well, in my tests I had a gps pps source locally which did not feed into
chrony, but was used only to look at the time offset of the system. Mind
you I used only one server, another PPS server running at that time
ntpd. Thus both systems were withing microseconds of UTC (as given by
the PPS) and the noise was thus all network noise. 


>
> This is a topic that needs serious "instrumenting".

Agreed. 

>
> H



More information about the questions mailing list