[ntp:questions] New to NTP on Linux

dusty bin junkmail at mmscomms.demon.co.uk
Wed Nov 3 10:32:37 UTC 2004


"David J Taylor" <david-taylor at invalid.com> wrote in message news:<2up5fhF2cg07uU1 at uni-berlin.de>...
> Dusty Bin wrote:
> []
> > The issue with both Linux and Windows is that the timekeeping is
> > rubbish (about +/- 80mS over periods of a few days, and time1 set to
> > around 0.6 seconds).
> 
> http://www.david-taylor.myby.co.uk/mrtg/daily_ntp.html
> 
> The top three graphs show what can be achieved with Windows, and the 
> bottom graph shows how some software can upset Windows 2000!  Yes, 80ms is 
> not good.

Thanks Dave and Brad...

I had a similar set up to Dave (actually 3 machines running w2k all
peering with each other and all sync'ed to the internet), and in that
configuration I obtain quite good results, typically within a couple
of mS unless there are other applications interferring with things -
exactly what Dave's link shows.
In this situation, I assume that the principal error is mainly the
computers ability to maintain a stable knowledge of time... so, if I
go from internet sync'ed to a reference clock (be it 1 microsecond for
Garmin or 100 nS accuracy for Oncore) the performance would be the
same, or marginally better (ref clock interrogated at 64 sec instead
of 1024 sec for internet servers).

The machine where I installed the NMEA ref clock isn't doing anything
else, that is why I am rather suspicious of the instability I have
seen.  In fact I needed to delete all the other peers/servers (except
.lcl.) from the configuration in order to force ntpd to use the NMEA
clock and so get a decent record of this clock's stability.

Hopefully someone will be able to enlighten me about the previous
questions on PPS, so that I can track down what is going on - as I
remain rather doubtful that the 1PPS signal is actually being used at
all at the moment...



More information about the questions mailing list