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

unruh unruh at wormhole.physics.ubc.ca
Tue Apr 19 20:19:49 UTC 2011

On 2011-04-19, David Woolley <david at ex.djwhome.demon.invalid> wrote:
> David J Taylor wrote:
>>  From chronyc I would need to be able to use a simple Perl script to 
>> extract the numbers to be plotted - such as the Offset in the graphs 
>> above.  An easy job if the format is standardised and machine readable.
> As I keep saying, offset does not accurately measure time keeping 
> accuracy.  A low offset can mean that you are tracking traffic 
> variations on the LAN, or interrupt latency variations, rather than 
> local clock frequency variations.

I would say offset does measure it. However, if what you want is the
rate, then you can also plot that. However, you could get a highly
non-varying rate by never changing it. The only way you know whether or
not your rate is right is by looking at the offsets. 
Note that it is ntpd that only looks at the offset. The rate is adjusted
to decrease the offsets (+offset and you increase the rate, - offset and
you decrease it). The way it gets "rid" of network fluctuations is by
making the correction smaller than you would need to get rid of the
offset by the next measurement. chrony fits the past, up to 64 data
points to estimate the rate. That averages out the network fluctuations
much more effectively than does ntpd.

> It is also a pessimistic estimate of accuracy when everything is locked 
> up and stable.

How is it a pessimistic estimate-- it is the only estimate you have. 

More information about the questions mailing list