[LinuxPPS] Some Info about the Project I'm working on

[Reg Clemens]

> Thank you all for your input. 
> 
> Hopefully the following will help clear up how the system is configured:
> 1) PPS - CPU's 1, 2, and 3 receive PPS via I/O pins on their parallel
> port
> 2) GPS - Only CPU 1 receives via COM1. 
> 3) Processor Communications
> 	a) CPU 1 - COM2 to  CPU 2 - COM1
> 	b) CPU 1 - COM3 to  CPU 3 - COM1
> 4) Accurate - As accurate as possible
> 
> I really like the idea of using ethernet to sync the clocks. To do so
> I'm thinking I'll incorporate a switch into the system. However, I'm
> concerned about how reliable messages sent using ethernet are. Isn't
> their some lag time and acknowledgment issues sending/receiving data via
> ethernet?
> 
> I truly appreciate all the help. 
> 
> Adam
> 
> On Fri, 2007-09-07 at 18:45 +0200, Rodolfo Giometti wrote:
> > On Fri, Sep 07, 2007 at 10:41:22AM -0600, clemens at dwf.com wrote:
> > 
> > > Yes, but my reading of his message was that he had both the GPS data
> > > and the PPS signal routed to all three machines, mabe Im wrong.
> > 
> > Nonono, you are right! I agree with you, it was just to fix it. :)
> > 

OK, as Rodolfo mentioned, you would need both the PPS signal and the
PPS data to run ntp on a given machine.  Since you only have the PPS
signal going to two of the machines, ntp wont run on them.  

Now as an aside, the ONCORE driver in ntp would in fact be ok with this
situation if you just split the data channel three ways, but most of the other
drivers (I think) would get hopelessly confused with the 'traffic' from the
other two ntp drivers...

That said, the TCP/IP solution should work really well in your case.
All three machines will be on the same stretch of ethernet cable, or on
the same switch, so the time for a message to go from A->B will be the
same as the time to go from B->A.  ntp takes this time into account in
setting the times of the two machines on the ethernet, and its only when
these time become long, or assymetrical (via strange routing on the 
external ethernet) that this is a problem.

your comment re:
   4) Accurate - As accurate as possible

still leaves a lot of questions open in my mind.
If you need miliseconds, no problem.
If you need microseconds, its going to be tough.
If you need nanoseconds, forget it.
    (or in this case the one crystal for all three machines plus some significant
     work to get things REALLY syncronized  would be necessary)

Good luck
-- 
                                        Reg.Clemens
                                        reg at dwf.com