2003-09-05T11:13:01 James Stevens: > > On sick, sick platforms this produces errors, > > I've never heard that before. Useful knowledge. I assumed all platforms > could run multiple child accepts.
The one I know about is Solaris 2.5.1 (fixed in 2.6, I
believe). Sick minds come in clutches, though, so even if there's
no Solaris 2.5.1 in your future, it's worth keeping the obscure
possibility in mind.
> > {Apache} Indeed, but it's solving a harder problem
>
> Yes, and no ... desktop users don't like having to wait to send their
> mail... if there aren't enough scanner processes available, then the
> user will be told "Try again later". They don't like that !
Ahh, this is a different deployment model.
I wouldn't dream of deploying something that's attempting a
near-impossible job, like MIME scanning, directly where clients can
hammer it.
Clients would submit to an MTA (I favour Postfix) which would
do whatever fast (constant-time!) front-end checks (anti-relay
at a minimum) they must and then enqueue the message and accept
committment for it. The client can go away. That front-end queue
drains through the hideous chain of disgusting hacks that try (and
sometimes succeed) to undo all the damage that Microsoft has done
to the world. Users wouldn't see their pacing, and the queue-drain
would manage the maximum concurrency that the mime hackers attempt
to run with.
-Bennett
pgp00000.pgp
Description: PGP signature
