[ntp:questions] What is the NTP recovery time from 16s step in GPSserver?

unruh unruh at invalid.ca
Thu Nov 1 16:20:14 UTC 2012

On 2012-11-01, David Woolley <david at ex.djwhome.demon.invalid> wrote:
> David Taylor wrote:
>> On 01/11/2012 08:40, David Woolley wrote:
>>> David Taylor wrote:
>>>> On a recent restart from cold, the server here took about 15 minutes
>>>> to get to within 250 microseconds, and about an hour to be within 10
>>>> microseconds.  Ultimately it is within about 3 microseconds.
>>> 3 microseconds of what?  How are you measuring the difference between
>>> software clock time and UTC, independently of NTP?
>> That's the offset as stated by NTP.
> That's not the time error.  Under ideal circumstances, it is several 
> times higher than the actual error, but with a low value like that, it 
> could easily be orders of magnitude worse (e.g. if there were interrupt 
> latencies in the 100s of microseconds and up, or, for network time, if 
> there were asymmetric delays in the milliseconds).

Interrupt latencies in my measurements tended to be at the one or 2
microsecond level. (drive a pin on the parallel port up, measuring when
you did that, and then measuring the interrupt output time and looking
at the difference.) Of course they are all one way, so averaging does
not make this error smaller.
When I once tried to cascade two interrupt service routines on the same
interrupt togetter, the second was about 10usec later than the first--
indicating the interrupt service time of the first. 

> At 250 microseconds, do they all have the same sign? The claim for 
> chrony would be that it gets to the state where the offsets are balanced 
> about zero, much faster.

More information about the questions mailing list