[ntpwg] Further to the timestamping issue

[TSG]

Rob I wanted to respond to this comment of yours.

----- Original Message ----- 
From: "Rob Seaman" <seaman at noao.edu>
To: <ntpwg at lists.ntp.org>
Sent: Wednesday, June 18, 2008 3:47 PM
Subject: Re: [ntpwg] Further to the timestamping issue


> TSG wrote:
>
>> it seems totally reasonable that in instances where there is an
>> embedded HW based timekeeping system like a 1588MC or similar, that
>> the NTP calls should be able to point to it rather than only getting
>> time off the local host-based OS Clock through the local NTP Service
>> Daemon.
>
> You appear to be talking about something entirely different than what
> Dave and PHK were discussing.  However...

No not really. But as to Leap Second management my response to you below 
should allow for that as well.

>
> Why should an external clock, locally attached, be believed more than
> an external clock attached to a remote server?

It shouldnt unless it is a master calibration resource like a properly 
calibrated 5071A for instance. The question is whether the remote clock is 
in fact a clock and not a reference oscillator.

For the same reason it would be believed remotely. Because there is 
calibration regimen which document's either remote clocks' operations.

> Either NTP is a way
> for two clocks to keep better time than either alone - or we should
> stop using it.

My point was whether certain REFCLOCK operations should be allowed to be the 
actual NTP Server's Time Source and not the TOD meaning that NTP becomes an 
interface to that HW system and a standardized protocol for interrogating 
that time-keeper device.

In my model the NTP Listener would query and sidestep the TOD register read 
through the REFCLOCK interface for devices which were more than reference 
oscillator's - i.e. that had a TOD calibrated into them from a CAL LAB.

This would allow for really stupid things like multiple sources of Time to 
be used inside the same systems (something probably foreign to you Rob but a 
really powerful idea for creating ensemble based timeservice models in a 
single system).

The idea that I was using a 5071A as an absolute source of time info means I 
want to query the reference device directly , not have some shared-use OS 
query it for me. Its why IEEE1588 is so powerful in an NTP world. In the 
best of all possible worlds, I want to be able to have three or more voices 
voting in an ensemble and a weighting model for their accuracy, jitter, and 
drift.

That means that NTP which becomes a single voice is a problem the way it is 
today.  What it needs is a way to abstract its timekeeping from the HOST OS 
TOD management so that the OS can be truely a client of the time-keeping 
process and not a client/facilitator of that process.

>
> Rob Seaman
> NOAO
>
> _______________________________________________
> ntpwg mailing list
> ntpwg at lists.ntp.org
> https://lists.ntp.org/mailman/listinfo/ntpwg