[ntp:questions] NTP vs chrony comparison (Was: oscillations in ntp clock synchronization)

David Woolley david at ex.djwhome.demon.co.uk.invalid
Sat Jan 26 09:55:05 UTC 2008

David L. Mills wrote:

> Well, I have done a market survey of sorts, if you can count my 
> consulting clients. There seems general agreement that 1 ms is a good 

Examples of specialized uses, rather than typical IT manager requirements.

> network on the sea floor off the Washington state coast. They need a 
> millisecond for experiments lasting months, not just 8-hour shifts, and 

The significance of the 8 hour shifts is that it means that start up 
transients are experienced every 8 hours of up time, not just 2 or 3 
times a year, so they are much more important.  (Also, down times are 
more like 60,000 seconds, and the hardware is commodity quality.)

> that when the experiment boxes get rather warm. Crony might work here as 
> well, but it would have to track large swings in temperature.

Nothing I've read so far indicates that chrony would have any problems. 
  In any case, I would have thought the thermal mass of the ocean would 
make temperature changes slow and therefore easy to track.

> I hear you say "100 ms" which I interpret as 100 milliseconds. Even 25 
> year old fuzzballs could to much better than that on the congested 
> ARPAnet. Did you mean 100 microseconds?

100ms is a rough estimate of the point at which almost any time 
synchronization method that has sub-second timestamps will work, with 
just very simple clock disciplines.  It represents a point at which ntpd 
is used because it is there, rather than because it is needed to meet 
the requirements.ent behaviour, as they could just use ntpdate in a cron 

I presume that the 100 microseconds is the true error, measured by out 
of band means, as my impression is that one should expect offsets in the 
low milliseconds on Unix, and maybe into the very low 10s on Windows.
(The last time I ran ntpd on Windows was some time ago, and a fairly old 
version, but I still believe that it doesn't like running on loaded 

More information about the questions mailing list