ietf
[Top] [All Lists]

Re: Predictable Internet Time

2017-01-03 10:33:33
I agree with most of what Patrik is saying. With one major difference, The
deviations file needs to be signed because it is a security issue.

And once you sign the file, the next question is... by whom.

I would much rather eliminate unnecessary trust dependencies than
perpetuate them.


Use of time zones at the protocol level is generally an option. XML Schema
uses UTC. Most implementations of IETF protocols use either UTC or UTC with
a numeric offset.

So it is an issue but mostly a UI issue rather than a security issue. It is
still necessary to have a signed table of the expected changes though
because agreeing to meet once a week at 10:00 London Time is not the same
as agreeing to meet once a week at 10:00 UTC+0 due to the fact that there
will be summer time half the year. A person in New York may attend meetings
in London and so the adjustments need to be known globally.

Contrary to what the calendar people tell me, these issues are not
'difficult'. What is difficult is using programs that make unfounded
assumptions about the adjustments to apply rather than letting these be set
by the user.



On Tue, Jan 3, 2017 at 1:29 AM, Patrik Fältström <paf(_at_)frobbit(_dot_)se> 
wrote:

There are a number of complicating factors here. One being that we have
multiple time scales. Another being that parties do believe things are what
they really are.

The main issue with smearing is that UNIX time is really not counting
number of seconds from the epoch, which might be what people think. If it
was, smearing would not be a good thing as the number of seconds from the
epoch ends up being one less due to the smear than the continuous series of
SI-Seconds. The number of seconds is guessing that each day have the same
number of seconds. So leap seconds are ignored.

Instead the goal with the smear is only to be correct according to UTC.

A correct implementation would of course be to have the number of seconds
from Jan 1 1970 be correct, and then a correct translation to UTC. Which
requires (both) knowledge of the number of leap seconds -- which you only
know for each 6 month interval (sort of). So one do not know the
translation between number of seconds since epoch and UTC if you look say 3
years forward in time.

I think personally, as long as we do have leap seconds:

- we should have the leap second information available somewhere in clear
machine readable format. Some suggestions exists, including encoding it in
A-records in DNS ;-)

- we should look at having the time since epoch really be the number of
SI-seconds since the epoch

- we should have translation between number of seconds and UTC take leap
seconds into account

- we should fix the code that do not accept 61 seconds in a minute

Now, we can also say we should stop having leap seconds, but I feel that
is a _different_ matter and different discussion. I am myself not clear
over what is the correct thing regarding leap seconds.

What I am sure of is that I think most of the problems we have is because
of bad programming (including in old UNIX days the priorities although
correct at the time have continued to let the time_t definition continue to
be wrong).

   Patrik

<Prev in Thread] Current Thread [Next in Thread>