[ntp:questions] Peers with and without PPS Source

Richard B. gilbert rgilbert88 at comcast.net
Mon Feb 12 22:47:39 UTC 2007

Thomas Kayser wrote:
> I have an interesting not really problem, but mystery.  I have 2 Debian Etch 
> machines set up with Garmin 18 GPSs and PPS (Debian1-PPS and Debian2-PPS).  I 
> have 2 other Debian Etch machines which have no PPS source directly attached 
> to them (Debian3-NOPPS and Debian4-NOPPS).  The Debian1-PPS and Debian2-PPS 
> use their own PPS source, and each other's PPS source, among other servers. 
> Debian3-NOPPS and Debian4-NOPPS use both the Debian1-PPS and Debian2-PPS, 
> among other servers.  On Debian1-PPS the offset for the Debian2-PPS time is 
> in about 4 us, and vice versa.  On the NOPPS machines, the offsets for the 
> PPS servers are 1-2 ms, and within us of each other.  
> So, it seems if a machine has a PPS source, other PPS sources used by that 
> machine will have much lower offsets than machines with no PPS source.  Is 
> this a fluke?  We can try to solve this mystery, hopefully for better 
> timekeeping for everyone.

I don't find this at all surprising!  1-2 milliseconds is good for 
transmitting time over a LAN.  The best I have seen over a 100 mbit 
switched LAN is about 400 microseconds.  This can only be achieved with 
a rock solid server using an atomic clock, GPS, or equivalent.  A server 
that is "clock hopping" among several good internet servers with 
slightly different notions of the correct time (as seen locally) cannot 
synchronize its clients this closely.

If you want really tight synchronization, feeding each machine a PPS 
signal (with careful attention to compensating for cable delays) is a 
good way to do it.

More information about the questions mailing list