[ntp:questions] Re: ntpdc vs iburst

David Woolley david at djwhome.demon.co.uk
Sat Nov 20 09:35:53 UTC 2004


In article <TrSdnYc0VZvz0ADcRVn-3Q at comcast.com>,
Richard B. Gilbert <rgilbert88 at comcast.net> wrote:
> Nagy Bela wrote:

> I don't see that using the local clock as a server gains you anything at 

I completely agree.  There is too much urban mythology about this feature.
Most people don't need it.

> >C) A simple ntpdate from ip-up.
> >
> This option simply sets your clock to approximately the correct time.  
> It does nothing to adjust the frequency of the clock oscillator; the 
> clock will continue to drift at its native rate.  You could be gaining 
> or losing two milliseconds per minute or worse.

If you have a system with the kernel time discipline code, one can
do quite well with ntpdate (and mustn't use ntpd -g instead) if one
calibrates out the static frequency error (measure the free run drift
over several days) and sets the kernel idea of the clock frequency to 
compensate.  In an air conditioned room, this is capable of about
30 seconds a year, even without occasional phase corrections.  At home
it is well under 1 second a week, and I generally only correct after
using Windows.

This does depend on your having a system that doesn't lose timer interrupts.

This is the relevant part of my startup script:

# time
/usr/sbin/tickadj 10000
/usr/sbin/ntptime -f 9.624
/usr/sbin/ntptime -s 0




More information about the questions mailing list