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

unruh unruh at wormhole.physics.ubc.ca
Mon Apr 18 04:25:54 UTC 2011

On 2011-04-18, C BlacK <rblak at non.net> wrote:
> Why would it take ntpd ten hours to achieve its accuracy?  Can this be explained in laymans terms and
> mathematically

Because ntpd does two things. It corrects the clock purely by looking at
the instantaneous offset error, and slowly changes the rate of the
compter clock to eliminate that error. It also throws away 80% of the
dat it collects in order to reduce the error due to assymetric travel
times between the computer and the remote time reference. The two
together means that ntpd takes a long time to fix say 10ms errors, or
30PPM rate errors. 

There is another program called chrony which operates much more rapidly
(minutes rather than hours) and more accurately (factors of about 3
better) than ntpd. 
It however runs only on Linux or Unix or BSD, not Windows. 
(It uses the past N measurement, where N can be up to 64, to determine
the rate and offset errors and then rapidly corrects them.) 
The reason that the chrony technique has not been applied to ntpd is
because Mills does not trust the technique. He has spent a lot of time
testing and making sure that ntpd is stable, and worries that chrony has
not been as throughly tested, despite its use without problems by many
many people. 

In the past it also did not have the refclock support, but it does now. 

>>> Absolutely normal!  NTPD can sometimes need up to ten hours to achieve 
>>> the accuracy it is capable of.

More information about the questions mailing list