Jan Hoevers wrote:
> Dennis Hilberg, Jr. wrote on 27-10-2007 20:40:
>> Right, but your system is only *accurate* to it's synchronization distance, 
>> which is the root dispersion + 1/2 the root delay.
>>
>> Your server, as viewed from my stratum one, has rootdelay=29.432 ms and 
>> rootdispersion=46.082 ms.  That's a synch distance of 60.798 ms, even though 
>> your reported offset in the same output is offset=-0.939 ms.
> 
> I'll have to do some reading on that first.
> 
> I've tested my system from two different locations over the internet 
> (and compared it with gps). It is certainly not 60 ms off.

No no, it's not 60 ms off, it's -0.939 ms off.

> I guess that's the maximum possible error, with these servers and 
> network paths. But again, I have to read a bit more about this first.

Yep, synch distance (root dispersion + 1/2 root delay) = maximum error, with 
respect to the root of the ntp chain.

> I didn't claim I can guarantee accuracy better than 1 ms, but skipping 
> the nanokernel and bringing timing granularity back to 1 ms would 
> degrade accuracy.

David Mills explains it a bit better in a post on the 
comp.protocols.time.ntp newsgroup on 8/30:

David L. Mills wrote:
 > Richard,
 >
 > For better or worse, the NTPv4 spec and the reference implmentation
 > define precision as the time to read the system clock currently a few
 > hundred nanoseconds. The resolution on the other hand is the smallest
 > increment that can be measured by the system clock, usually one
 > nanosecond with Unix timespec structures. Your interptetation of the
 > root dispersion is not pecisely correct. What is correct is that the
 > root synchronization distance, equal to the root dispersion plus half
 > the root delay, is th maximum errot bound.
 >
 > Dave
 >
 > Richard B. Gilbert wrote:
 >> Dennis Hilberg Jr wrote:
 >>
 >>> Jason Rabel wrote:
 >>>
 >>>> Those numbers seem much more reasonable.
 >>>>
 >>>> The root dispersion is the maximum error between the local clock and
 >>>> the
 >>>> root of the NTP chain. Your time on each system is accurate to
 >>>> within the
 >>>> range given above.
 >>>
 >>>
 >>>
 >>> My problem is that I misunderstood you when you said 'precision'.  I
 >>> was simply referring to offsets grabbed from ntpq.  Perhaps Guy was too.
 >>>
 >>
 >> The NTP world uses the word "precision" to designate the value of the
 >> low order bit kept by the system clock; e.g. the smallest possible
 >> difference between the two consecutive time stamps.  It's generally
 >> specified as a negative power of two.
 >>
 >> My Sun Solaris (SPARC) boxes have a precision of -21 but an accuracy
 >> of plus/minus 5 milliseconds when synchronized from the internet.

Sorry, I should have worded that differently.

-- 
Dennis Hilberg, Jr.      [EMAIL PROTECTED]
NTP Server Information:  http://saturn.dennishilberg.com/ntp.php
_______________________________________________
timekeepers mailing list
[email protected]
https://fortytwo.ch/mailman/cgi-bin/listinfo/timekeepers

Reply via email to