[ntp:questions] Re: PPS Output from NTP

Ulrich Windl Ulrich.Windl at RZ.Uni-Regensburg.DE
Mon May 9 13:05:05 UTC 2005

"Tony" <anthony.flavin at bt.com> writes:

> Having a fairly large NTP deployment, I would like to perform a
> hardware
> measurement of one of the NTP servers (Stratum 3) against the PPS
> signal
> from the GPS on the Stratum 1. However I can't think of a way of
> getting the
> Linux ST3 to generate a pulse in User mode that won't get delayed by
> the
> scheduler etc and introduce errors to the readings.
> The PPS kit would be an option as this does not run in user mode, but
> will
> only echo an incoming PPS.
> Does anybody know of a way to get a the kernel to generate a 1PPS
> signal
> that would not suffer this problem?

IMHO you'd have to modify the kernel to emit both, a pulse at the closest
timer tninterrupt to the new second, and an offset value describing how far
the pulse actually was from the real new second. In User Land you could try to
read the time and busy-poll until the start of a new second, and then emit the pulse.

> Someone must have done this, or else how do we really know we have the
> time
> correct?

What about this:

server -----------> client
 |                   |
 |                   |
 v                   |
judge <---------------

The judge can tell you what the relative offsets are.


More information about the questions mailing list