[ntp:questions] 500ppm - is it too small?

Richard B. Gilbert rgilbert88 at comcast.net
Mon Aug 17 21:27:39 UTC 2009


nemo_outis wrote:
> "Richard B. Gilbert" <rgilbert88 at comcast.net> wrote in
> news:RfWdneXv8elEAhTXnZ2dnUVZ_sKdnZ2d at giganews.com: 
> 
>> nemo_outis wrote:
>>> "Richard B. Gilbert" <rgilbert88 at comcast.net> wrote in
>>> news:poydnd1spod4pBTXnZ2dnUVZ_vudnZ2d at giganews.com: 
>>>
>>>> I still haven't seen any argument that supporting a value greater
>>>> than 500 PPM is worth the trouble!
>>> I still haven't seen any evidence that there is much trouble - only 
>>> anticipatory speculative whinging. 
>>>
>>> Nor, for that matter, has there been much other than speculation and
>>> vague anecdotal recollections that the 500 ppm limit *may* have had
>>> some justification and not been entirely arbitrary.
>>>
>>> This doesn't inspire much confdence about the documentation of the 
>>> architecture and design decisions regarding ntp.
>>>
>>> Regards,
>>>
>>>
>> I think "Das Buch" a/k/a  "Computer Network Time Synchronization: The 
>> Network Time Protocol" by David L. Mills might help you understand the
>> design and how it was derived.
>>
>> See:
>> http://www.amazon.com/s/ref=nb_ss?url=search-alias%3Daps&field-keywords
>> =NTP&x=0&y=0 
>>
>> I think you will need some background in "control systems", "phase 
>> locked loops" and other such esoterica.  I don't pretend to understand
>> the math.  I simply use the software derived from it to synchronize my
>> clocks.
>>
>> NTPD does that nicely!
>>
> 
> I have the book and I've read it.  And although I am a mechanical 
> engineer, not electrical, I have a solid background in control theory and 
> I do understand the math.  
> 
> The justification for 500 ppm in "Das Buch" is very thin gruel and 
> amounts to litle more than "argument by repeated assertion."  In fact, 
> the book itself discusses a 1997 survey (section 6.6) that shows time 
> offsets of hundreds of milliseconds, so far from being unusual, are 
> common. This (inter alia) makes the book's light regard for large offsets 
> seem more than a little cavalier.
>
I don't know where 100 millisecond offsets might come from unless 
something is badly broken or a server is extremely "ill chosen"; e.g.
if you are in New York and using a server in Tokyo, one might question 
both your sanity and the accuracy of the time you are getting.
Offsets of the order of 10 milliseconds are far more usual.

As for the 500 PPM limit, it seems to work.  Its theoretical 
justification may be weak but "it works" counts for a lot!




More information about the questions mailing list