From: "Samatason Ltd" <chris@xxxxxxxxxxxxxxx>
Hi
What's the best approach to filtering out spam at server mailbox level? I am
running FC5 and SendMail...
Best Regards
SpamAssassin can perform exceptionally well. Under almost any conditions
it can perform quite well.
So let's ask about your environment. How many people are you serving?
If there are "many", are they a relatively homogeneous group? Are you
contemplating using your own smtp server? (Modern spam filtering has
made this a nice experience on receipt of email and a little more
difficult for administering the send side of the world. These days
you must be careful about your rDNS and SPF setups for best anti-
spam performance.)
Personally I am filtering for a sharply limited number of people whose
interests are not that far apart. I happen to use fetchmail to capture
mail so some of the nice toys like greylisting are not in the cards.
I feed the incoming messages from fetchmail (or for awhile when
postfix was in the batch from postfix's incoming mail) to procmail.
It's old. It's crufty. Nobody likes it. It's arcane. It works MY way.
That made the difference. I've faced all temptations to run AmavisD,
MailScanner, various milters, or any of the other MDA tools floating
around. Running SpamAssassin from procmail is "falling off a log"
simple unless your mail system's security and "jail" gets in the
way. You can learn how for most of the MDAs out there from the
SpamAssassin WIKI. If you dig back for old messages from me on this
list with SpamAssassin mentioned inside you'll find one or two long
descriptions of my setup. That'd save me typing it again, although
I do have some new techniques I've added to my setup to handle "open"
mailing lists that do not filter spam all that effectively. That has
set me up for 99.99% spam catch and about 99.999% correct ham
classification. I diddled scores on per use BAYES rules to get this.
I also use just a whole lot of the SpamAssassin Rules Emporium's
rule sets. They make at least that last digit difference on catching
the spam. I'm also setup to hand train spam and ham per user, too.
Although if the automatic training starts off right it can do quite
astoundingly well. It can also go wrong if some bad training happens
early on.
{^_^} Joanne
