[ntp:questions] NTP in a Linux cluster

Unruh unruh-spam at physics.ubc.ca
Tue Sep 8 15:12:07 UTC 2009

"Richard B. Gilbert" <rgilbert88 at comcast.net> writes:

>I don't know if servers or peers can be given their own minpoll and 
>maxpoll.  There is very little reason to meddle with the default MINPOLL 
>and MAXPOLL!  NTPD adjusts the value in use as needed.

>A somewhat oversimplified explanation is that a short poll interval is 
>used to correct large errors quickly and the long poll interval allows 
>NTPD to measure and correct small errors very accurately.

Well, no. It is better at correcting small DRIFT errors accurately. It
does worse on offset errors. But then again, drift errors are largely
irrelevant if the machine is checking itself frequently. It is if it
goes off air for a while that drift errors become nasty.

>I would try to avoid using servers on the Internet.  You can get much 
>better stability and accuracy using a hardware reference clock.  This 
>can be a GPS receiver, a LORAN receiver, an "atomic clock" or, perhaps, 
>one or two other devices.  Symmetricom and Meinberg Funkuhren both offer 
>equipment that you may find useful.  It's expensive but if you need 
>stable and accurate time it well be worth the expense.

It depends entirely on his requirements. He never did say what his error
tolerance is.

>If you are able to site an antenna where it will have a view of most or 
>all of the sky, a GPS timing receiver is a very good choice.  The 
>antenna is a little bigger than a quarter and smaller than a half 
>dollar.  Mine sits on top of a "Leaf Guard" rain gutter on the Southern 
>side of the house.  It works like a champ!

His problem is not the antenna, but the wiring needed to connect that
antenna to his system. 


More information about the questions mailing list