[ntp:questions] Getting precision when server side only offersseconds... Ideas?

sgb1010 at hotmail.com sgb1010 at hotmail.com
Tue Feb 26 12:52:34 UTC 2008

Hi Jason,

On Feb 25, 12:26 pm, ja... at extremeoverclocking.com (Jason Rabel)
> Soleth,
> > Explaining it better, it is a kind of network switch for which I don't
> > have total access,
> > and can only talk through a protocol, that offers me a date in format
> > YYYY-MM-DD hh:mm:ss. Period.
> Does the server auto-output the info "on-time" every second, or only
> dynamically reply when requested by the client? Some hardware that I have
> seen has an option to start an auto-output where each line of time is
> printed within so many ms of the zero mark.

Unfortunately no. It replies when requested.

> > Am I right and is it impossible to get the desired precision or does
> > anyone have
> > any idea of how could I accomplish this? Is it possible to perhaps
> > "guess" the
> > millisecond "behind" the seconds I receive? The satellite delay is
> > something on
> > the order of 700-1200 ms.
> With the high latency of a satellite link it might be hard to pound it into
> submission (querying multiple times per second until the second changes).
> You can account for the bulk of the delay just in the round-trip time. If
> you wanted to dig deeper you could figure out the distance to the satellite
> and the delay caused by that, then delay caused by the hardware, etc...
> At the same time you could be querying approx once a second (either more or
> less often depending on latency) and as the second changes you can slowly
> converge on the correct time by adjusting your query time to hover around
> the time-change.

Yes, that's what I am working on right now. With the time I sent the
request and
the time I received the reply recorded, I know the total time. I can
then make an
estimate on the round trip time + hardware delay involved.

That, of course, if the hardware delay is not too high, in which case
it will screw
things up. =)

With that in hands, I will see if I can try to guess the "milliseconds
behind the
seconds" by requesting it several times, and seeing if I can converge
or at
least reduce the error. Working on a test to see if this is doable.


More information about the questions mailing list