[ntp:hackers] Precise synchronisation on a local network

Hal Murray hmurray at suespammers.org
Thu Jul 6 07:27:09 UTC 2006


>   - we buy lots of PCs and cannot rely on their clock to be high
> quality (some  have 10ppm clocks, other 100ppm ones) 

What do you mean by "10ppm clocks, other 100ppm ones"?  Is that NTP's drift 
value, or a number printed on the crystal?  (or found in the manufacturer's 
data sheet)

I think that crystals accurate to 10ppm will be too expensive to use in 
normal PCs.  Are you using fancy/expensive industrial PCs?


> To improve the accuracy of the NTP client we have  :
>   - set minpoll and maxpoll to 4 (16s) to accelerate convergence and
> increase  the number of measurement performed 

What happens if you don't do that?

I'm not surprised that causes problems, but I'm not sure I can explain it.

Do you know about the Allan Deviation?  The idea is that if you measure your 
clock over short intervals (seconds), the main error is in the measurements.  
If you measure over long intervals (days) the main error is the clock 
changing between measurements.  If you plot percentage error vs measuring 
time, you get a V shape.  You (probably) want NTP's polling interval to be at 
the bottom of that V.  That's probably many hundreds of seconds.  Maybe more.

One complication is the temperature of your crystal.  How stable is the 
temperature of your environment?  What is the load like on your PCs?

My room temperature changes by 15 F over a day.  That's 2.5 ppm, but it's 
slow enough that NTP tracks it.  On the other hand, a mostly idle system gets 
a big jolt of activity at 4 AM when a cron job does a lot of work and heats 
things up: >0.5 ppm over 15 min.  That's a lot harder to track.

Good reading here:
  NTP temperature compensation
  http://www.ijs.si/time/temp-compensation/
(Good background.  I'm not suggesting you implement it.)


Have you turned on logging?  What happens if you plot drift over time?


>  - we have few clients (2 or 4 PCs querying 2 servers)

That's a small number.  Are they close together?  Can you connect them all to 
a GPS clock?  (All you need is a RS-232 splitter box and some cables.)

Does your GPS clock put out a clean signal?  (Do you have a good antenna 
location?)




-- 
The suespammers.org mail server is located in California.  So are all my
other mailboxes.  Please do not send unsolicited bulk e-mail or unsolicited
commercial e-mail to my suespammers.org address or any of my other addresses.
These are my opinions, not necessarily my employer's.  I hate spam.





More information about the hackers mailing list