In message <4E5FAE64(_dot_)2060809(_at_)pscs(_dot_)co(_dot_)uk>, Paul Smith
writes:
On 01/09/2011 15:35, Menachem Moystoviz wrote:
Suppose one person (Let's call her Alice) wants to send an email to
another (Let's call him Bob).
Now suppose Bob's server is online for 4 hours every day.
According to section 4.5.4.1, Alice's server (or whatever server is
closest to Bob)
should retry sending the mail every half hour or so for 4-5 days,
meaning that Bob will eventually get the mail.
However, experience with Google's Gmail service has showed me that
usually, the server gives up after several seconds.
Is my sample biased or is this the norm, in violation of the RFC?
Moreover, will Bob receive the mail in my example?
Because this would mean that any user who turns his SMTP server on for
a couple of hours each day would receive his mail.
Well, the sender should retry, but there is no rule about how often or
for how long it should retry, just a vague suggestion (not even 'SHOULD')
If gmail gives up after a few seconds, then I'm a bit dubious. I'd
expect it to try longer than that, I'd have thought for at least a day
(unless it gets a permanent error report back from somewhere).
I'd say that most probably try for 3 days (gives admins chance to fix
their server after a weekend if it broke on the Friday evening). Busy
ones like gmail could try less because they have so much mail, and don't
care as it's a free service. As long as they generate an NDR it's
compliant with the standard.
However, if a mail server is online for only a few hours a day, then I'd
seriously suggest getting a backup mail server to accept messages during
the downtime (or putting the main mail server somewhere else with a
better connection), unless you don't really care about getting mail...
And poll when coming online using TURN after authenticating.
Also if you do this you should ensure that your backup MX can reject
non valid addresses or you will generate a lot of backscatter.
There are a number of different techniques that can be used to do
this.
Mark
For a single MTA I'd say that it has to be online for 'pretty much all
the time'. Offline for an hour or two for maintenance over a few weeks
should be OK in my experience (eg 99.5% uptime).
--
Mark Andrews, ISC
1 Seymour St., Dundas Valley, NSW 2117, Australia
PHONE: +61 2 9871 4742 INTERNET: marka(_at_)isc(_dot_)org