[ntp:hackers] Further to the timestamping issue

David L. Mills mills at udel.edu
Tue Jun 17 01:46:23 UTC 2008


The the maximum error statistic is defined as half the root delay plus 
the root distpersion. The root disperision is the error due to the 
maximum credible frequency error of the primary reference source 
multipled by the time since last synchronized by that source. Additional 
minor statistics are added as defined in the specificiation. It may be 
your opinion that a maximum error statistic is not userful, and you are 
welcome to that opinion.

Yes, I did juju magic and managed to cram three 32-bit words into two 
leap bits. But, you have to look further in the documentation to find 
out how I did that. You might start by looking for the leapseconds file 
and, should the mystery about how other machines find out about it, you 
might enjoy the autokey specification. The leapseconds extension field 
shouldn't require autokey. Gotta fix that someday.


Poul-Henning Kamp wrote:

> In message <485679D6.3010407 at udel.edu>, "David L. Mills" writes:
>> Poul-Henning Kamp wrote:
>>> I did not say they were not correct, I said they were of no use.
>> If other drivers don't follow the intended semantics, that's not the
>> fault of the semantics.
> I'm not saying it is, but I'm advocating adding some metrics with
> semantics people can understand and use.
> Even if they were correctly implemented, the root delay and root
> dispersion is of no use to NTP users.
>> The leapseconds values include the epoch of the next/last leap,
> I fail to fully appreciate how you compressed that into the
> two bit field of the NTPv4 protocol.
> Or are you talking about some undocumented extension field ?

More information about the hackers mailing list