[ntp:questions] Trading jitter for offset

Unruh unruh-spam at physics.ubc.ca
Sat Mar 28 21:25:45 UTC 2009


"David J Taylor" <david-taylor at blueyonder.neither-this-bit.nor-this.co.uk> writes:

>David Woolley wrote:
>> David J Taylor wrote:
>>
>>>
>>> Is there a parameter I can tune from the ntp.conf which would reduce
>>> the offset, at the expense of increased jitter?  Perhaps I'm asking
>>> for the impossible?
>>>
>>
>> Reducing minpoll will help, but you are basically observing an aspect
>> of why Bill Unruh advocates chrony.  ntpd has poor startup and
>> frequency change transient handling.

>Thanks, David, I though that might be the case.  The systems are already 
>running quite a small minpoll, and I'm not sure I want to reduce it 
>further (to be kinder to the servers involved).

>It's a pity that someone has not ported chrony to Windows, as that's the 
>platform where I need it.  I guess they might need to add some of the 
>interpolation code as well.  An end-node, client-only port (no ref-clock) 
>would be fine.

chrony does not support refclocks anyway. The key problem is in the clock
discipline. chronymakes extensive use of the adjtimex system call in linux
which allow adjustment of the clock frequency (what is the size of the
interpolation adjustment between timer ticks), of the tick size (ie how
many nsec should be added to the clock each time the timer tick occurs)
and the adjustment timeout (Ie I tell it I want it to adjust the time by
3.11ms and it will "rapidly " slew the clock to do that.
It does not have ntp's own adjustment meechanism ( once a second adjust the
freq of the clock).
Ie, the main problem I think is that you would have to mimic that
technology for windows. 
Curnow had an intrest at one time in porting to windows, but his interest
or time to keep up chrony is low now so it will have to be done by others.
 




More information about the questions mailing list