[ntp:questions] ntp server help

David Woolley david at ex.djwhome.demon.co.uk.invalid
Sat Aug 9 14:40:21 UTC 2008


Unruh wrote:
> mikel at irontec.com (Mikel Jimenez) writes:
> 
> 
> And you are using ntpdate why? Why not set up ntp on the system and use it
> to discipline your system's clock. Note whether you will get 1/100 sec
> accuracy depends on the network etc. 

Although it isn't particularly clear, I believe he is running ntpdate on 
  a machine that is properly synchronised, to measure the offsets of the 
cameras.

Of course, the error in the cameras may well be an order of magnitude 
better than the offsets measured, either that way, or the more approved 
way of using configuring the cameras as ignored servers on the 
synchronizing machine.

However, my gut feeling is that this application required that it be an 
explicit requirement to buy the cameras that they maintain an offset 
less than 10ms.  If that was the case, then he should be pursuing this 
with the vendors of the cameras.  If not, he should put it down to 
experience.

It seems unlikely to me that the cameras are runing NTP, rather than a 
basic implementation of SNTP, and even if they are running NTP, it is 
unlikely that they have included any of the recent optimisations for 
fast convergence, so he should be waiting several hours, after powering 
up the cameras, before doing anything critical.

It's also worth noting that one really needs a local radio reference 
clock (although I suspect he's actually using the local clock as 
reference) to guarantee 10ms accuracy on the server.  It is possible to 
get this most of the time with an internet time reference, but some care 
is needed to achieve it all the time.
> 
> And why are you comparing the offset? 

He wants the video feeds from the cameras to have their RTP timestamps 
within 20ms (probably he meant 10ms) of each other.  I'm not sure why, 
as it would seem easier to retime outside of the camera.  I'm guessing 
the RTP bit.
> 
>




More information about the questions mailing list