exim / amavis-ng overload

[David Fokkema <dfokkema@ileos.nl>]

Hi group,

Had my system offline for about 36 hours to try out fai (on a different
drive). After this period, sunday evening, already late, I reconnected
my original drive, started the server and watched the connection coming
up. Many, many mails were downloaded (thanks to this fine group, :-) and
almost immediately, my server became irresponsive. I managed to squeeze
in and run top which came on after three minutes and saw my load
increase from 20 to 30 to 50 to 77.81. I had a few exim processes and
_lots_ of amavis processes running. My server started killing processes
(out of memory) and I decided to power down (using the button, nothing
else worked). I think I know what went wrong: there was no limit of
running exim processes. Each process started amavis to scan for viruses
and my server came to a dead stop. I disabled amavis and my mail came
right through. Apparently, delivery _is_ limited because I had only one
spamassassin running from my user's procmail at a time.

Two (groups of) questions:

1. What can I do when my server overloads like this? Logging in was
   _very_ difficult or not possible (timeout of 60 seconds on login
   prompt...)

2. How do I limit the number of exim processes? Exim is run from inetd
   and the infopages showed me that smtp_accept_max will not work. Is
   this indeed a problem? Should I run exim as a separate daemon and
   enable smtp_accept_max (or keep it at default 20, better than no
   limit...)? Is every server which runs exim from inetd vulnerable to
   this (whether deliberate or not) DoS attack?

Thanks!

David