[ntp:questions] Re: Generating a PPS output.

Alexander Fisher alex at alexfisher.me.uk
Sat May 27 15:37:17 UTC 2006


David J Taylor wrote:
> Alexander Fisher wrote:
>> Hello
>>
>> I have an application where I'll have 2 PCs on an ethernet LAN.  One
>> of these will be fed a GPS PPS input and act as a level-1 NTP server.
>> The other PC will be an NTP client.  I need to get the client PC to
>> output a PPS signal.  The new PPS signal needs to match the original
>> to within 1 ms.   The LAN will be 100Mb/sec or better, won't contain
>> many switches and will hopefully not be highly loaded.
> []
>> Many thanks,
>> Alex
> 
> It would obviously be far simpler to run a second PPS line from the 
> original PPS device, it's just RS-232 levels after all.  I presume that 
> this is not possible for your environment?

That's correct.  I've been told that replacing the existing cabling is a
*very* expensive option.

> There was a Microsoft research paper quoted here recently showing that a 
> specially configured NTP could get within (IIRC) a few tens of 
> microseconds for playing audio streams.  Maybe some tips from that 
> article?

http://research.microsoft.com/users/misard/papers/aes2004.pdf

Looks interesting.  But I don't really like the idea of significantly
altering NTP.  What if it ended up costing more than replacing the
existing cabling!  I'd be quite happy with ten times less accuracy than
what is suggested in the paper.

>From what I understand, the key to getting sub ms NTP accuracy is very
low and symmetric network latency.  I've then just got to figure out how
to not lose any more accuracy when generating the PPS out.

Kind Regards,
Alex



More information about the questions mailing list