Re: synching to atomic clock software

From: Gavin Eadie (gavin@umich.edu)
Date: Thu Mar 31 2011 - 20:43:09 UTC


First, I should say I'm not familiar with what software Windows uses for time-keeping, so my comment are generic.  The 'standard' network time protocol (NTP) software, that is the Official Reference Implementation, claims a "few millisecond" accuracy on unix systems with a decent connection to the Internet.  It operates, as Paul described, by constantly pulling the system clock into synchrony with time references on the internet.  Multiple time servers are used and multiple samples of those sources are accumulated -- the worse servers are eliminated from the process and the remaining "truechimers" used to adjust the computer's clock.  A good description is at:

  http://www.eecis.udel.edu/%7emills/exec.html

Pretty much all unix-based computers (including Macs running Mac OS X) use the Reference Implementation of NTP software.  However, how that software is configured can make a big difference to the accuracy of the time derived.  For Mac OS X, for example, you can only specify one time server in the appropriate preference panel and that one server is placed in the NTP configuration file -- defeating the sophisticated averaging and culling of multiple servers which leads to the most accurate time.

It sounds like Tardis performs its duties in the same way as NTP (maybe, at its heart, it is NTP).  
On my laptop, NTP resets my clock by about a quarter of a second about four times a day.
All the above doesn't take into account the quality of the software that is using the clock -- NTP operates at a low level and takes network and computer latency into account, but if your application gets interrupted or is just plain slow, all bets are off!

For people who find this a fascinating topic:

   http://www.ntp.org/

There's also a little more conversation on this topic from mid-Feb, this year, on this list .. Gavin

On Mar 30, 2011, at 11:57 PM, Paul Grace wrote:

> Look for Tardis, it can be set to step the time or change clock frequency
> depending on the error, and will talk to multiple time servers.  I run it
> under Win7-64 without issues.  I also have it running on my weather computer
> which has been on for months at a time, and has converged on good accuracy
> (sub-second per day) which then serves the time to all my other computers on
> my LAN, reducing latency errors all around.

_______________________________________________
Seesat-l mailing list
http://mailman.satobs.org/mailman/listinfo/seesat-l



This archive was generated by hypermail 2b29 : Thu Mar 31 2011 - 20:44:43 UTC