[ntp:questions] SNTP accuracy

E-Mail Sent to this address will be added to the BlackLists Null at BlackList.Anitech-Systems.invalid
Thu Dec 27 06:02:30 UTC 2012


Ran Shalit wrote:
> I read in http://www.sematech.org/docubase/document/4736aeng.pdf
>  that "SNTP provides accuracy typically within 100 ms".
> I would like to ask if this means that on a specific
>  system combining two machines connected by ethernet (LAN),
>  it can be any number *changing in run-time* within 100 msec
>  or that we will reach more or less a constant delay number.
> This is very important because in case it is a constant number,
>  even 98msec delay, we can always delete this delay time (98msec)
>  and get a time synchronization around zero. Is that right ?

Thy don't you try it?
 It couldn't take more than a few minutes to setup SNTP;
  then let it run for a few hours while gathering data.


The actual drift will never be constant, however your Perceived Drift
 would depend on your equipment and the resolution of your measure.

If you are measuring in increments of 100ms, probably.

If you are measuring in increments of 1ms, not likely on a windows PC;
 although probable on a BSD PC or embeded RTOS device
  and likely on Linux PC / Device.


The computer's oscillator(s) will drift, probably mostly with temperature,
 although there will likely be a innate bias in one direction too.
 {The oscillators that drive the computers timers / counters / clocks}

 Replacing the oscillator with a TCXO / CXO / Rubidium reference oscillator,
  might get you better than micro-seconds.

 NTPd would likely do a much better job of minimizing drift than SNTP
  with the factory oscillator(S), still your choice of hardware platform,
  and operating system could make a significant difference.

 NTPd with a external reference clock, like a GPS with 1 PPS
  or 1 PPS Rubidium reference, might get you better than micro-seconds.


-- 
E-Mail Sent to this address <BlackList at Anitech-Systems.com>
  will be added to the BlackLists.



More information about the questions mailing list