[ntp:questions] A faster settling NTP

David Mills mills at udel.edu
Tue Dec 22 20:51:02 UTC 2009


It already does. If you start ntpd without a frequency file, it will 
measure the intrinsic clock frequency error over a fifteen-minute period 
and set the clock within 1 PPM. Alternatively, if you set minpoll as low 
as 8, it will converge 8 times faster than the default. It will, 
however, be 8 times more vulnerable to network delay spikes common on 
our department Ethernet switch. The choice of fifteen minutes is 
arbitrary based on the 1-PPM tolerance; other values are possible.

 Folks sometimes fail to see this issue in context. Ordinarily, it 
happens only when ntpd is started for the first time. By defaults, and  
assuming the frequency compensation has stabilized within 1 PPM, a step 
transient up to 125 ms decays and crosses zero in about 3000 s, 
overshoots 6-8 percent, then decays to less than one percent in a few hours.


David J Taylor wrote:

>"Richard B. Gilbert" <> wrote in message 
>news:wfKdnRhdnr8nvazWnZ2dnUVZ_judnZ2d at giganews.com...
>>unruh wrote:
>>>Nonsense. chrony does it, without loss of accuracy (chrony is about 3
>>>times as accurate as ntp is) or stability. It will correct a few 
>>>second initial error in far less time than ntp takes for a .01 sec 
>>>and without stepping.
>>Then why don't you use chrony and stop bugging us?  If it can replace 
>>NTPD under most common scenarios for normal and emergency operation and 
>>do a better job, I'm sure that it will eventually replace NTPD.  Does 
>>anyone see that happening yet?
>I would be quite happy to try chrony were it not for the fact that it 
>doesn't work with Windows and it doesn't support reference clocks.
>I would be even happier if there were a version of NTP available which, as 
>an option, had a fast convergence algorithm for its initial operation, and 
>switched to the existing NTP algorithm once initial stability had been 
>questions mailing list
>questions at lists.ntp.org

More information about the questions mailing list