[ntp:questions] Re: Laptop

David Woolley david at djwhome.demon.co.uk
Mon Mar 13 21:34:21 UTC 2006

In article <4414CFF0.7030900 at ntp.isc.org>,
mayer at ntp.isc.org (Danny Mayer) wrote:
> Gerriet M. Denkmann wrote:

> ntpd was designed at a time that Internet access was unreliable and you
> would be dialing the NIST number in Colorado, US, to get an accurate
> time source. So yes, this is actually normal.

I seem to remember that version 3 handled dialup servers specially and 
didn't use the normal phase locked loop.  (Maybe it was just that the
clock driver did its own projections.)

> Set up your server with iburst options and don't use any restrict lines.
> in the startup line add -g which will allow it to reset you clock even
> if the clock is way off.

This is going to give him a very poor frequency estimate.  With infrequent
polling it tends to be best to lock in the frequency and then periodically
just correct the phase.  The dialup server code used the whole period 
between polls as its frequency baseline.  (The only way I know of locking
in the frequency is to use ntpdate for the updates.)

> ntpd works fine. Just restart ntpd every time you connect as the IP
> address of your local system will change and ntpd does not yet handle
> dynamic address changes on the local machine.

It's by no means certain that the address will change.  If he is using
broadband and DHCP is being operated correctly, it shouldn't change
(reportedly many cheap ISPs break this deliberately).  He could also
have a static address, and, he may be connecting the office network
using ethernet, with properly operating DHCP.  So, whilst one can say
the address will probably change, we don't know this for certain.

> > iBook G4 with Mac OS X 10.4.5

Laptops can also go into suspend states, in which the CPU clock is not
being used at all and the real time clock is used to interpolate - this
can cause havoc for ntpd.  I don't know if OS X goes to a full suspend.

> > /etc/ntp.conf = "server time.asia.apple.com minpoll 12 maxpoll 17"

This will probably never acquire lock, and if it ever does, will soon
lose it.  If it does acquire, you will need to be connected for more
than four hours continuously before it does acquire, even if the 
frequency and phase are dead centre.

> > /var/run/ntp.drift = "0.00"

To stand any chance of acquiring lock with the above minpoll, you will
need to have this correct to rather better than 10 ppm.  Having it 
present, but with the wrong value is about the worst thing you can
do here.

More information about the questions mailing list