MailScanner vs. SpamKiller

[Peter Bonivart]

Max Kipness wrote:
> I’m not doing Bayes at the moment as it seems to be a real hassle doing 
> the training.

I don't know for other sites but I don't bother with training. I use the 
  default autolearn feature (<0.1 ham, >12 spam) and it works great. If 
you have Exchange on the inside it's not easy to get a correct mail to 
learn from.

> So my question is what can I do to improve the whole system? What 
> tweaks? Will DCC help out a lot? Are there any better RBLs? Tweaks to 
> SpamAssassin?

Definitely turn on Bayes, it will help a lot even without additional 
training. Also use DCC, it's a really good design. It's easy to install, 
fast and stable.

Here are my top SA traps:

SpamAssassin    369,153
...HTML_MESSAGE 290,859
...BAYES_99     285,210
...RCVD_IN_BL_SPAMCOP_NET       255,030
...DCC_CHECK    232,846

Out of a total of 369,153 messages identified as spam Bayes was 99% sure 
that 285,210 were spam, that's 5.4 points right there. The best RBL for 
me, as you can see, is spamcop and right after that follows the DCC 
checks. HTML_MESSAGE is a low scoring test that doesn't affect the total 
much.

-- 
/Peter Bonivart

--Unix lovers do it in the Sun

Sun Fire V210, Solaris 9, Sendmail 8.12.10, MailScanner 4.25-14,
SpamAssassin 2.63 + DCC 1.2.30, ClamAV 0.67 + GMP 4.1.2