On Thu, 11 Apr 2002, Jesus Cea Avion wrote:

> >   At this point, I think the scalability of database mail solutions is
> > unproven one way or the other.

There are Oracle-based email solutions. They basically suck, but this
isn't an Oracle problem - it's because the coders writing the "Mail
Transports" involved are less skilled than an infinite bunch of monkeys
with typewriters pounding out Shakespearean sonnets and haven't bothered
reading the applicable RFCs before starting (and this goes for Dan
Bernstein's Qmail too).

Hint: RFC "MUST" = if you don't do this, I'm going to come pound on your ass

Hint2: RFC "SHOULD" = if you don't do this, I'm going to come pound on your ass.

Where "I" = a large bunch of admins who are pissed off with
broken-by-design mail transports (such as qmail) causing DoS attacks on
our systems with massive parallel connects, or causing our systems to
become bogged down with bogus bounces (Qmail, Groupwise, Notes,
Exchange, PostOffice, etc)  or causing our systems to back up because
the destination system has suddenly decided it will only accept 2
messages per hour from anyone (Exchange, Notes, Groupwise, etc)


My personal opinion after evaluating database storage of several tens of
millions of variable (but highly similar and short length) message blobs
vs hashed filesystem storage is that databases are _not_ suited for this
kind of activity.

If you're running NT or DOS based systems then this may be different,
but for almost all *nix filesystems it holds true.

Most of Qpopper's problems stem from trying to cope with people leaving
mail on the mail server, simultaneous access from differing IPs or
processes (local vs pop3 access, etc)

The POP3 protocol is unsuitable for this as it doesn't support switching
folders, forcing users to leave everything in the inbox, which then
grows to massive size. This is because POP3 wasn't designed with these
possibilities in mind. It was and is a _simple_ access method, for a
world where access methods are ever-increasingly complex - but only for
about 0.1% of the users.

If users are doing these things or want to do these things, use IMAP.
We'd be switching to it in droves if it was renamed POP4...


I'm _not_ slamming Qpopper. It's very good at what it does, but it's not
a swiss army knife and we really need to resist the urge to turn it into
one.

AB





Reply via email to