[ntp:questions] Using Trimble TSIP under Linux
david-taylor at blueyonder.co.uk.invalid
Tue Nov 27 08:10:46 UTC 2012
On 26/11/2012 23:59, David Woolley wrote:
> David Taylor wrote:
>> server reached over the LAN to have an offset of -0.028 milliseconds,
>> whereas its own PPS source is reported to have an offset of -0.001
>> milliseconds. Is that the right sign and amount for the offsets you
>> might expect comparing the LAN input and the GPIO pin interrupt for
>> the PPS source?
> Offsets should be randomly distributed between + and - signs. If they
> are all of one sign, you are not locked up or there is a bug. They
> should be of s similar magnitude to the jitter.
> They may not be evenly distributed, because of limitations of the
> network, but they must be spread across zero.
David, for clarification:
On the card which is synced purely to LAN servers, over a day the mean
offset of the FreeBSD stratum-1 server (the selected sync source) is
0.000 ms, and that of two Windows stratum-1 servers is 0.014 ms. The
mean jitter is 0.041 ms (FreeBSD) and 0.048 and 0.280 ms (Windows). The
offset of WAN-based pool servers was in the region of 2 - 8 ms.
On the card with the PPS source, the mean offset of the other servers is
-0.020 ms (FreeBSD) and -0.008 ms (both Windows). Mean jitter is 0.230
ms (FreeBSD) and 0.036 ms and 0.230 and 0.047 ms (Windows). The offset
of WAN-based pool servers was in the region of 2 - 14 ms.
The WAN is via a cable modem connection, rather asymmetrical at 30 Mb/s
down and 3 Mb/s up.
Looking at the graph, the offsets of the LAN servers are very stable,
and much less variable than the WAN servers. They also are clustered
around zero. I conclude that there's no bug and the NTP is locked.
What I may well be seeing is the differences in interrupt latency
between the different hardware and operating systems. It's not a
problem for me.
More information about the questions