Ariën Huisken, 09.11.2006 13:03:
> Hello list,
> 
> I have several boxes running TSL2.2 with postfix/amavis/spamasassin/clamav.
> 
> In the last months the amount of received spam is very very high. So  
> high, everyone is complaning about it.

> Now I was playing with the settings (tag levels) in amavisd.conf, but  
> setting the kill level below 4 causes normal email to go in to the  
> filter, while a lot of spam still comes through.

Hard to give specific advise without knowing your setup in detail so 
I'll just describe my setup:

* Well trained bayes with over 30k spam and 60k ham active
* I upgraded amavisd-new to latest version and added p0f-support
* Added extra score to messages coming from Korea and China
* Added RBL checks against Intersil, uribl blacklist and uribl greylist
* Greylisting for 10 minutes, as some bots started resending after 5 minutes

P0f support means passively checking the senders packets to determine 
OS. I then score Windows 2000 and XP host a little since there ought to 
be no SMTP-server on XP. p0f also tries to determine how many hops away 
the sender is and I can subtract a small amount of points of the sender 
is close.

The next step I see is new postfix to get greet-pause, something I did 
experiements with a year ago and shows same potential as greylisting.

I guess I will be filing a bug-request to have amavisd-new upgraded and 
postfix. I just have to take a little closer look at the implications 
first. Upgrading amavisd-new using the TSL srpm was a bit tricky, but 
not much.

Matthias mentions primary/secondary MX'es. Most spammers hits the MX 
with highest priority-value in hopes of that server having less 
spamchecks. This makes sense where companies have commercial solutions 
and has to pay per server for their antispam...
Some suggest a fake tertiary MX which do not respond at all, another 
suggestion is to give a slighte increase in the score if message was 
delivered through a tertiary MX.

But bayes ought to be very effective for you seeing that you 
(apparantly) are Dutch. Most spam is English. This makes at least bayes 
extremely accurate in Norway.

I have this on 2 servers. My personal, which is the volunteer tester, 
and my company MX. The company MX has a kill-threshold of 9 points and 
we have very little, if any, spam coming through. I go through the 
quarantine with some regexes and similar to try to catch false positives 
but haven't found any yet (after 2 years). Home server kills at 4.21 
points. I reached that value after grep'ing and analyzing the average 
scores I was seeing.

My biggest advise is greylisting, to be honest:
http://www.tyldum.com/spam-year.png
This is my home server. Blue is total emails, green is the number of 
spam detected. Guess when I started greylisting?
_______________________________________________
tsl-discuss mailing list
[email protected]
http://lists.trustix.org/mailman/listinfo/tsl-discuss

Reply via email to