[ntp:questions] Re: How much Jitter can commercial NTP tolerate (I'm talking a huge jitter here)

Terje Mathisen terje.mathisen at hda.hydro.com
Thu Mar 31 08:17:35 UTC 2005

candice.uhlir at ngc.com wrote:

> I have a need ('please don't ask why...to synchronize a computer system
> that runs  on an aperiodic clock tick of 64ms where that "tick" can be
> stretched out to as much a 700ms at certain times.  This does not hurt
> this one of a kind system since it just clocks off of these pulses.
> OUr problem is that our data is "time tagged" with a count of how many
> of these 64 pulses have arrived since a reference time of 0 and we try
> to process these data elements on the ground.  We can't speed up 64ms
> clock to real time....however if we can slow our ground system to run
> on these stretched out 64ms pulses all should, in theory, be ok.
>   We were wondering what would happen if we removed our GPS stimuli
> from our NTP server (A Datachron) and replaced the gps stimuli with a
> time generated from our 64ms wierdo time pulses.  What would the
> commercial NTP software in a SUN or SGI workstation do when faced with
> this abnormally huge jitter??  ANybody have any other guesses on other
> effects we may see?  

I'm not sure, but I can make an educated guess:

NTP is designed to have a maximum capture range of +/- 500 ppm, i.e. 
maximum 1.8 seconds/hour.

When you get a single pulse lasting 700 ms, ntp will either give up 
totally, or assume it just dropped about 11 pulses.

I think the best you can do is to generate a totally separate 
application clock, consisting of just the current count of pulses 
received, and then run your full processing loop once per tick.


- <Terje.Mathisen at hda.hydro.com>
"almost all programming can be viewed as an exercise in caching"

More information about the questions mailing list