On Fri, 11 Oct 2002 19:11:42 +0100 Steve Haslam <steve@???> wrote:
> On Sat, Oct 12, 2002 at 12:48:40AM +0800, Suresh Ramasubramanian wrote:
> > rwelty@??? (Richard Welty) writes:
> >
> > > VERP is kind of a trade off. i like the solution in majordomo 2,
> which is
> > > to send out VERP mailings a couple times a month, which is enough to
> do
> > > some reasonable list cleaning.
> >
> > Yup - massively parallelized deliveries, as compared to tcp slowstart
> and
> > nagel penalties paid - which rather tends to mount up?
> Why do I get the feeling you two are talking about totally different
> things?
we're not, actually.
VERP is (as i understand it) the product of djb, author of qmail. it most
naturally fits in qmail style delivery, where multiple RCPT TO: targets
are disdained and individual email messages are sent regardless of the
opportunity to batch up multiple destinations in the RCPT TO:
it happens to be the only rational way to address bounces automagically
that i'm aware of; the MAIL FROM: address is crafted to encode the
recipient, so when the bounce comes back, you can figure out what address
caused it. however, of necessity it forces qmail style delivery tactics.
qmail style delivery can be argued 'til the cows come home. it presumes
that there are no limits on bandwidth between email servers, and no
difficulties connecting to destination MX hosts.
additionally, it forces you to eat tcp startup costs for every individual
destination address, whereas multiple RCPT TO: destinations allow you to
spread that cost over many target addresses for a given destination.
which is why i like the majordomo 2 compromise -- VERP occasionally so you
can clean the bad addresses, but most of the time use multiple RCPT TO: to
get the most of out of those hotmail.com connections that are so difficult
to come by.
richard
--
Richard Welty rwelty@???
Averill Park Networking 518-573-7592
Unix, Linux, IP Network Engineering, Security
