[ntp:questions] NTPD can take 10 hours to achieve stability

Chris Albertson albertson.chris at gmail.com
Mon Apr 18 18:36:10 UTC 2011


On Mon, Apr 18, 2011 at 9:28 AM, Rob <nomail at example.com> wrote:

> You assume that your measurement scale is very coarse.  Your reference
> is seconds and you can only measure plus or minus one second.
> But that is not true in NTP with a local reference clock.  It can measure
> to microseconds or even nanoseconds.

The measurement scale is always  very coarse compared to desired precision.

That same applies if the scale is seconds or nanoseconds.  If you can
measure to one part in N and desire the rate be set to one part in M
and N<M then you most wait for time equal to or greater than to M/N.

There is another layer over this,  your reference clock might have som
uncertainty or jitter on it.  In the case of NTP the 1PPS signal
causes an interrupt and the local clock is used to time stamp the pps
event.  Neither the local clock nor the a pps is perfect so one must
average to reduce the noise.  I think NTP does this by using a long
time constant on the loop, in other words by not adjusting the rate
quickly

The OP asked for a simple answer about why it takes so long to do
something so seemingly simple as setting a clock.  The answer was that
you should expect it to because you always have to wait to see it a
clock is running at the desired rate.  The more precision  desired the
longer you need to wait.

-- 
=====
Chris Albertson
Redondo Beach, California



More information about the questions mailing list