On Fri Feb 29, 2008 at 20:08:41 -0000, Vic wrote:
> > Normally when i build an email server I have a recipient accept list, that
> > is, any name not on the list gets bounced by Postfix, and thus NOT
> > processed by spamassasin etc.  If this was in place on Monday, there would
> > not have been a DoS as it was only 11,000 spams during the day, which I
> > consider not too bad.
> 
> There is a school of thought that says that's the way to do it - if you
> reject by bad addresses first, you leak your valid address list to
> spammers. But I don't subscribe to that - check for valid recipients
> first, or your load goes through the roof...
  I think it is obvious that dropping bad mail should be done as
 soon as it is possible to do so, precisely to avoid becoming victim
 to a DOS.
  Right now I'm running a spam filtering service and the general
 order of tests is:
    1.  Connecting IP or hostname based rejection.
    2.  HELO tests.
    3.  Early talker tests.
    4.  Valid user tests.
    ...
    ...
        anti-virus test
        anti-spam test
  (After testing I discovered that testing for viruses was faster
 than testing for spam.  Having said that I'm only seeing .4% viral
 mail, so it might make sense to reverse the order - that way I dont
 invoke ClamAV at all for spam mail.)
  Current volume of mail is in the region of 200,000 messages a
 day and I think if I didn't do the username testing early I'd not
 be able to handle that much without much more load.
Steve
-- 
http://mail-scanning.com/