[ntp:hackers] Does ntpd need to whine more ?

David L. Mills mills at udel.edu
Mon Oct 3 15:38:33 UTC 2005


You didn't read my message. The requirement to use a crafted discipline 
algorithm applies only to an intermediate server with upstream sources 
and downstream clients. Symmetricom knows this as I have personally told 
them so. There is no requirement for Symmetricom to adopt this algorithm 
in the case of precision sources such as GPS or cesium or even ACTS, as 
some Symmetricom producs use.

The NTP clock discipline algorithm was not designed to discipline 
precision sources directly; however, in my experience with external PPS 
sources, both the kernel discipline and daemon discipline work quite 
well. See pogo.udel.edu or rackety.udel.edu, both of which hold the PPS 
in general to within a few microseconds. I don't think it makes a lot of 
sense to do much better than that for a modern computer network; but, if 
the jitter and wander could be better controlled, there is no reason why 
the existing discipline could split the nanosecond.

I don't understand your comment about the discipline "wandering around" 
with a precision source. It doesn't do that, a fact that can be quickly 
confirmed using the simulator. You are invited to confirm yourself and I 
would be really interested in any exceptions you may find.

There is no intent to outlaw research on clock discipline algorithms, 
only a requirement that the closed-loop transient response in all 
servers and clients be compatible and stable over the poll interval 
range, and this requires the poles and zeros to be carefully managed, 
especially if multiple strata are involved. Think of the response of a 
whip to a transient, and then consider the response of a whip to another 

Your tone is dissapointing, but I am not going to argue with you. 
Please, and I am serious about this, take your reservations to the NTP 
WG and be heard.


Poul-Henning Kamp wrote:

> In message <43414333.4070001 at udel.edu>, "David L. Mills" writes:
>> Be advised the NTP WG will likely include the ntpd clock discipline as a
>> required component of a conforming intermediate server that synchronizes
>> to an upstram server and provides synchronization to a downstream
>> client.
> That would not only be a bad idea, it would also be downright stupid
> because it would essentially outlaw research and development on this
> very topic.
> And it would make it impossible for companies like Symmetricom to
> deliver a conforming server because their timebases are waaaaay
> better than anything the NTP clock discipline knows how to handle.
> (If you run the NTP clock discipline on a OCXO or better timebase,
> it wanders aimlessly around for ages before it realizes that it has
> very little to do and it never ever gets close the necessary
> timeconstants.)
>> This is necessary for stability in a distributed, multi-stratum
>> subnet.
> That is rubbish.
> The telcos SOnet networks that span the globe have solved far tricker
> synchronization problems with a huge mix of different algorithms
> and have managed to create a stable network despite many different
> manufactureres blunders over time.
>> If you have other agendae, whou should particiapte in the
>> discussions and make your opinions known.
> My opinions are herewith on the record.
> I do not have the economic resources to participate in IETF WGs, sorry.
> Poul-Henning

More information about the hackers mailing list