Load Balancing with Postfix [and SpamAssassin]

Michael Torrie torriem at chem.byu.edu
Wed Jan 18 23:36:52 MST 2006


On Wed, 2006-01-18 at 21:43 -0700, Andrew McNabb wrote:
> Fortunately dspam is trying to allow you to combine together several
> different learning methods (and they're experimenting on additional
> ones).
> 
> Anyway, there's my rant for the day. :)

Well, my point was if you can use multiple means of controlling spam,
starting with the least-cost and most-effective first (graylisting,
dspam, and then spamassassin), you can actually catch even more spam
(maybe 99.9999%) at way less the price (resource-wise) than running just
one of those filters (say spamassassin, but probably not dspam alone).
I know they tell you that you can't get something for nothing, but it
turns out there are techniques you can do to increase effectiveness and
lower resource costs if you are smart about it.

For me, I don't have a huge spam problem yet, but the little spam I do
get does clog my linode server quite a bit (for just a few minutes at a
time).  I certainly know that a full-on spam dump would bring my server
to its knees in a matter of minutes (it's only an 80 MB RAM linode
machine).  I don't have a corpus large enough to make dspam effective at
the moment.  So I still have to rely on spamassassin (which works great
for me).  If I could just reduce the amount of spam hitting spamassassin
it would continue to work well into the future.  So for me, implementing
gray listing will discourage 75-90% of the spammers from even connecting
to my mail server, and the rest can be taken care of by spamassassin
relatively cheaply.  I'm surprised that gray-listing is so effective,
and surprised it's not talked about a lot more.  Anyone have any
negative experiences with it (besides not being able to get instant e-
mails from a new source, like a mailing list confirmation)?

Michael


Michael

> 
> /*
> PLUG: http://plug.org, #utah on irc.freenode.net
> Unsubscribe: http://plug.org/mailman/options/plug
> Don't fear the penguin.
> */



More information about the PLUG mailing list