[calsify] Timezone Service Protocol not quite complete

[Steve Crocker <[e-mail]>]

I was very pleased to see the work on the Timezone Service Protocol.  I have no argument about the format of the data elements or actions, but the timing model is incomplete.  One of the motivations is make sure accurate timezone data is distributed in a timely fashion.  In order to make good on this, we need to say more about the timing.  As noted in the anecdote cited below, time zone changes are sometimes abrupt.  How much notice is required in order to make sure time zone changes are guaranteed to be propagated to all devices?  And on the other end of the system, how frequently do devices need to poll or otherwise be notified of changes?

Historically, IETF protocols have tended to be a bit vague about timing parameters and it's been left to operational practice.  The results are frequently messy.  We have a chance for this one to be much clearer.  The timezone database now maintained by IANA is a pretty solid solution for the source of reliable, authoritative information, but there's not been much written other than anecdotes about how often the data changes and how much lead time is required to distribute the information.

In the present draft, the only wording I could find is at the section 4.1.3:

> Clients SHOULD NOT poll for such changes too frequently, typically once a day ought to be sufficient.  See Section 8 on expected client and server behavior regarding high request rates.

and the cautions in Section 8.

I don't have an axe to grind regarding the design of the protocol, but I will note that if the timezone data were published via DNS, both the throttling problem and the model of propagation time would be dealt with automatically.  In any case, even if there's no desire to distribute this data over DNS, I do believe we need to provide much stronger statements about the expected time to propagate changes and to be clearer about how the design implements such statements.

Steve




On Oct 30, 2013, at 11:23 AM, Cyrus Daboo <[e-mail]> wrote:

> Hi [e-mail],
> 
> --On October 30, 2013 at 7:52:55 AM -0700 [e-mail] wrote:
> 
>>>> Which argues for including the local time when the event will take
>>>> place. If you want to mandate that the time also appear in UTC, great,
>>>> but it's quite important that the local time also be there.
>> 
>>> And local time should be normative to deal with such unforeseen
>>> circumstances as a papal visit.  From the southamerica file in the TZ
>>> database:
>> 
>>> # From Daniel C. Sobral (1998-02-12):
>>> # In 1997, the DS began on October 6. The stated reason was that
>>> # because international television networks ignored Brazil's policy on
>>> # DS, they bought the wrong times on satellite for coverage of Pope's
>>> # visit. This year, the ending date of DS was postponed to March 1
>>> # to help dealing with the shortages of electric power.
>> 
>>> ;-)
>> 
>> Good point. I agree.
> 
> Timezones can (and do) change at very short notice in different parts of the world. Typically computer systems have timezone data cached as part of the OS, and only get updates when the OS is itself updated - often long after a timezone change has occurred or long after future events were booked but are now out of sync for participants in different timezones. To address that a number of folks in the calendaring and scheduling community have been working om a timezone service protocol - <https://datatracker.ietf.org/doc/draft-douglass-timezone-service/> - our initial focus for that has been iCalendar (RFC5545) based VTIMEZONE component delivery to calendar and scheduling clients and servers. However, we would like to also deliver OS-style timezone data to devices to break out of the requirement for those to be updated only when the OS itself updates. For unix-based OS's that means being able to deliver "raw" zoneinfo" data over that protocol.
> 
> Anyway, if you are interested in that work please comment over on the ietf-calsify list (cc'd) - we (the authors of that draft) - intend to get it moving to last call soon. There are already several implementations that have undergone testing at the Calendaring and Scheduling Consortium (CalConnect) interop events, so we are happy with it, but would appreciate more feedback.
> 
> -- 
> Cyrus Daboo
>