[ntp:questions] Win2k3 Server as NTP server?

David J Taylor david-taylor at blueyonder.not-this-bit.nor-this-part.co.uk.invalid
Mon Dec 7 08:47:18 UTC 2009

"David Woolley" <david at ex.djwhome.demon.invalid> wrote in message 
news:hfiddi$qc4$1 at news.eternal-september.org...
> I think he was really asking about the parameters that you log and how 
> you get them from ntpd.
> He is using ntpq to fetch the "offset" value.  If ntpd is working well, 
> this will be noise like and centred around zero.  Under those 
> circumstances, its jitter should be several times worse than the true 
> jitter in the software clock.  Windows applications' idea of th time 
> will be compromised by the limitations of Windows, in particular its 
> inability to interpolate between ticks,  There may also be a systematic 
> error due to network limitations, etc.
> If the offset is not noise like, the offset may be close to the actual 
> clock error, as ntpd has poor transient response to things like 
> temperature changes.

I hope the answer is covered in the Web pages to which I pointed, but if 
not, I use ntpq to get offset and sometimes jitter, and I enable the 
statistics logging with lines in the ntp.conf such as:

enable stats
statsdir "C:\Tools\NTP\etc\"
statistics loopstats

The files can be read by both Meinberg's and my own plotting programs.

I find that the offset is far from noise-like on some systems (Feenix and 
Narvik, both running XP), reflecting the temperature or rate of change of 
temperature.  On other systems it is more noise-like (PCs Stamsund and 
Hydra) and the loopstats averaged jitter on those systems is Hydra 1ms and 
Stamsund 25us.  Hydra is a Windows 7 system with LAN sync, and Stamsund a 
Windows-7 system with a GPS/PPS local reference.


It puzzles me somewhat why the same ntpd.exe running on Windows-XP 
has a much worse /apparent/ offset than one running under Windows-7.  Both 
PCs have a GPS/PPS reference, 16s poll, and both behaved similarly under 
Windows-XP.  In Windows-7, the interpolation scheme is disabled, which 
seems to result in a much lower average offset, but with an increased 
noise on the plots.  The Windows-XP machine shows an averaged jitter in 
the loopstats of 2.5us, compared to 25us for the un-interpolated Windows-7 


More information about the questions mailing list