[ntp:questions] So what? (Re: 500ppm - is it too small?)

Unruh unruh-spam at physics.ubc.ca
Thu Aug 20 05:40:21 UTC 2009


Nero Imhard <nim at pipe.nl> writes:

>nemo_outis wrote:

>> A clock with a frequency deviation is NOT, per se, broken.  If the
>> deviation, even if it be, say, 5000 ppm, is consistent, stable and
>> regular, then the clock could be a superb timekeeper.  

>Yes of course, but you miss the point (or at least my point). Such a 
>large deviation can easily be corrected for *before* ntp gets to sync 
>the clock. Again, the margin into which you should tune your clock rate 
>is a matter of taste and quite arbitrary.

And you miss the point-- ntp itself should be doing that. It is a
program that is supposed to discipline the clock. It should not ask you
to do part of its job.



>But since you are pushing it, here's my personal take: I deem it 
>reasonable to require a clock's frequency error to be in the same 
>ballpark as (let's say at most one order of magnitude larger than) its 
>instability. Anything larger should be (and is in fact easily) tuned out 
>through other means than ntp's clock discipline. 500 is *huge* already.

>I'm not versed in the theory enough to make any assertions about how a 
>large frequency error affects behaviour and performance of the clock 
>disciplining algorithm, but my gut feeling tells me it's better to have 
>a small frequency error. If someone who does know these things 
>convincingly tells me otherwise, I'll gladly adjust my opinion. For the 
>moment, I trust my gut.

>N




More information about the questions mailing list