Vladimir

Thanks for the links.

What I see is replacing the crystal on a PC-based NTP server with an OCXO.
I believe that most people are already using Rb clocks in NTP servers.

The question of interest is what is possible in the client.
We have run a lot of 1588 tests with various crystals and various network 
scenarios,
with both HW and SW timestamps, and with a proprietary clock recovery 
algorithm. 
Based on these tests I think that maintaining 1 us for realistic networks and 
pure SW timestamps
is not realistic, even when the client has a decent local oscillator.

I will try running a few tests with standard NTPv4 and report to the list.

Y(J)S

-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of 
Vladimir Smotlacha
Sent: Friday, April 01, 2011 11:43
To: [email protected]
Subject: [TICTOC] NTP server oscillator issue

Hello,

I'd like to contribute to the discussion about improving computer clock 
accuracy, that started yesterday at TICTOC meeting.

According to our experiments and evaluation of our NTP servers, very 
important issue is computer oscillator stability. I replaced the quartz 
14.318 MHz at mainboard by a OCXO. We use this hacks in more than 10 
servers, both NTP and Time-stamp authorities. We reached system clock 
accuracy few microseconds (PPS disciplined server) resp. less than 100 
microseconds (Stratum-2 with primary NTP in the same LAN). Note: it is 
objective time offset measurement outside the system. Offset reported in 
  NTP log loopstats provides better numbers.

Some of these results are described in Cesnet technical reports

http://www.cesnet.cz/doc/techzpravy/2007/ntp-server/
  and
http://www.cesnet.cz/doc/techzpravy/2008/time-stamp-authority/


regards,

        Vladimir Smotlacha
        


_______________________________________________
TICTOC mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/tictoc
_______________________________________________
TICTOC mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/tictoc

Reply via email to