[Top] [All Lists]

Re: 2821bis consideration - New 2nd attempt Retry Strategy recommendation

2007-11-18 17:32:48

--On Sunday, 18 November, 2007 16:14 -0500 "Robert A. Rosenberg"
<hal9001(_at_)panix(_dot_)com> wrote:

At 03:08 -0500 on 11/18/2007, John C Klensin wrote about Re:
2821bis consideration - New 2nd attempt Retry Strategy :

And that is exactly why the guidance in the spec says what it


My comment was specifically directed to the suggestion to trim
the give-up period back to 2 days. I pointed out that weekends
are not the only problem with 2 days. If someone DOES decide
that 2 days is enough, that should be defined as 2 business
days and automatically adjusted when 2 business days spans
more than 2 calender days. IOW: Account for Holidays and
Weekends that fall during the next 2 calender days (ie: Set
the give-up date on a message to 2 calender days after
submission). This still is US Centric about Holidays but it
does allow for the shortened give-up period when it will not
run into problems.

Sorry, I am rarely too succinct for people, but it happens.

Since I try to avoid being US Centric, and the protocol
certainly should not be, what I was trying to suggest was that
it is probably a bad idea to try to figure out and adjust for
"business days".  It is probably an exaggeration, but it often
seems as if every day is a holiday somewhere.  Since Friday is
the rest/ business break day in significant parts of the work,
any sort of international scheme has to consider that "weekends"
are three days long (Friday-Sunday) even before allowing for
differences that occur due to time zones and the date line.  

The consequence of this is that planning for "two business days"
will take one fairly close to a four calendar day window even
without allowing for national holidays.  If one does allow for
national holidays in a non-trivial number of countries, but
ignores multi-day holidays, one quickly gets to four calendar
days as well (although one then also has to sort out weekends
and non-weekends).

"Four" is in the spec because it is a plausible compromise
without getting bogged down in the details of multiple countries
and cultures and because, if one does get that involved, one
will often come up with four anyway.

I think there is one important way in which one might be
selective and that is on notification messages: it is just my
personal opinion, but I'm a lot more willing to dump a
notification message that cannot be delivered after a couple of
days than I am to lose a message that was sent out by a known
legitimate user or via an authorized relay operation.  For the
latter, I consider the specified four days to be a fairly bare
minimum for robustness.   

Again, just my opinion.


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