Re: [Ntp] CLOCK_TAI (was NTPv5: big picture)

[Philip Prindeville <[e-mail]>]

> On Jan 5, 2021, at 8:51 AM, Magnus Danielson <[e-mail]> wrote:
> 
> I lean towards pushing the conversion requirement onto the NTP source
> nodes just to achieve the intended cleanness of the rest of the
> infrastructure.


I’m going to push back against that.

I think that many protocols embed human-readable time notation unnecessarily.  Take the syslog protocol.  Is there a reason why it couldn’t just have used TAI time in seconds and milliseconds (or microseconds), and left it to some browser to do the rendering or “presentation” into a human-preferred format?

Take video security cameras.  Do they need to know that a particular timestamp corresponds to a minute/day/month/year/leap-second?  I don’t think so.  I think the browser there, too, could do the conversion.

Has anyone ever tried to run a “make” on a large code base just as a second leaped?  It’s a mess.  A lot of things get rebuilt unnecessarily.  And that’s with forward leaps.  If we ever need to shorten a day, I can’t begin to image how badly things will break.

This wouldn’t happen if we let machines deal with a representation of time that’s optimal for them (i.e. simple and unambiguous) and held off on conversion to human-readable format until it was actually necessary.

As a postmaster for 30 odd years, I’ve had to struggle with Received: lines having local dates and times that require me to pay attention to the UTC offsets to be aware of when DST took place (if it took place at all) for each timezone that a server lived in as it handled a message.  It’s a mess.  RFC-822 should have mandated UT or Zulu time (as that was the best we had at the time).

Or having logs sent to me, or logging into remote machines, and having to know the timezone of the provenance of those logs, then doing the math in my head to match that up with other events elsewhere.

Simple, unique, unambiguous, and universal time is what machines need.

Conversion to something that’s convenient for humans should be deferred until absolutely necessary (and it may never be!).