[ntp:hackers] Problems with Oncore???

[Bjorn Gabrielsson]

Terje Mathisen <terje.mathisen at hda.hydro.com> writes:

> bg at lysator.liu.se wrote:
> > I have a small box converting TTL <-> RS232 with a standard
> > MAX232. This conversion adds ca 1.5us delay. 
> 
> OK, that's in the proper ballpark, and similar to what I expect from my
> old TAPR interface board.
> > 
> > However I between my old Jupiter and a Trimble Acutime2000 (palisade)
> > I also see a difference of 13us, like Terje have with different
> > equipment. I have not fully explored which one is of by 13us in my system.
> 
> See my previous reply to Poul-Henning: This is what you get when you
> have PPS signals on two different serial ports on the same PC!

Except that the palisade driver does not use the PPS signal. From
driver29.html; "The Palisade's external event input with 40 nanosecond
resolution is utilized by the Palisade NTP driver for asynchronous
precision time transfer."  


> This means that the first (probably COM2 which is at IRQ 3) serial port
> to be handled will timestamp the interrupt, and only then can it go on
> to the other (COM1 at IRQ 4) signal. This gives a consisten bias for the
>  lower priority port.
> 
> Oncore has a perfect fix for this, in that the actual time of the PPS
> signal can be offset an arbitrary amount from the real top of the second.
> 
> By setting it way off, at 1 ms, the two tasks won't ever interfere, and
> I get pretty much identical performance from both sources. :-)

Was under the impression that multiple PPS-sources was allowed to a
single computer? Or is it only one PPS source that can be fed to the
kernel PLL?

--
        Björn