[ntp:questions] Synching to a Legacy System as an External Clock Reference

michael.antinucci at lmco.com michael.antinucci at lmco.com
Fri May 18 11:45:06 UTC 2007

> Not if that old legacy system isn't an architecture that ntpd has been
> ported to.

That is exactly the case. The old legacy system cannot be modifed.
There are discussions on upgrading the old system to use NTP, but we
can't hold our breath on that.

I did not mention this in my previous post, the systems do NOT have to
synched to a standard time zone or "real time" say UTC/GMT. The
systems I am trying to link together just need accurate time
synchronization to each other only, beginning from when the systems
start up.

> Use a daemon that listens to the UDP clock pulses and, after suitable
> preprocessing, sends them to ntpd via the SHM driver.

That is the approach I had in mind. What worries me is the "suitble
preprocessing" part as well as figuring out the exact parameters to
pass to the ntpd via the SHM driver. Does the SHM require a 1HZ PPS or
is my ~0.1HZ  pulse OK to use?

> The stability of the time base is of primary importance.

I imagine it is quite stable. The old legacy system is a very large
scale miltary system used for high-precesion data loading of weapon
data. I am in a prototype/feasiblity mode right now and it is
difficult (i.e. very time consuming) to obtain  the documentation for
the old legacy system that has the specs on the interface. I am hoping
David Mills is reading this post because (according to his web site)
he has a current contract working on the same system I am trying to
link to!

I really would like to synch my internal network to a external clock
reference (say an Internet site or GPS unit) that is very stable so I
can measure just how stable/precise/accurate the old legacy system is.
Unfortunaley, that is not possible in the lab environment I am working
in. So, since the single board computer I am using for the ntpd server
has a cheap clock in it that I know drifts in a big way, I am unsure
how to measure the accuracy of the old legacy system. The only tool I
have right now is an 8 channel digital oscillocope that I trigger with
general purpose output pins. Some of the nodes on my system trigger
those pins at absolute times and I can measure the jitter between them
and the master ntpd node. I am seeing no more than +/- 1 ms jitter
between the slave nodes and master node.

> Actually, it sends a packet containing the clock counter (ostensibly
> milliseconds since system startup) approximately every ten seconds.
> That's not the same thing as "just tick[ing] periodically".

That is correct.

More information about the questions mailing list