Anthony Metcalf wrote: > Hi All, > I have been hunting around on the web now for some weeks, trying to > find an answer to my question. I think I am incapable of being > succinct enough to get an answer through searching though. :) > > If there is a howto out there, or some other source that will answer > this question, please point me directly at that. > > Now for the question. > > I have a mail gateway, that intercepts mail to and from my company, > and virus/spam filters it, before passing it on to the internal > (Exchange) server or the internet. > > This system works, in that all mail is being spam and virus filtered. > I can see this from headers in the mails, and subject tags added by > spam assassin. > > The system is no tagging much though, on one mailbox in particular, it > tags maybe 10% of the mail. This is a web visible mail box with about > 99% spam. > > As the server my mail client interacts with is not the one spam > filtering, I would like to set up two accounts on the gateway box, ham > and spam, so I can farward mail to those accounts, and have > spamassassin learn what is ham and spam from them. > > How do I go about this? > > System Steup: > > Internet<-->gateway(linux,postfix, amavisd, clamav, f-prot, > spamassassin)<-->internal server(winodows, exchange) > > Thanks again for any help, and sorry for the long post.
Anthony, I have the same setup. As such, I would suggest doing a little tuning to spamassassin before worrying too much about training bayes. Although bayes works pretty well here, it is definitely not one of my biggest hitters. Plus in SA 3.0+ the scores for bayes have gone down quite a bit anyway so you're most likely not going to see a huge improvement in overall spam hit rates even if they all hit Bayes_99. My installation hit 90%+ of the spams "out-of-the-box" and am now up to ~97% with very little tweaking. I'm not even using any custom rules yet. With you only hitting 10%, it appears to me that you may have some configuration problems. I can definitely help you with the learning question as I do train my bayes from outlook clients, but like I said, concentrate on getting a better hit rate without bayes first. Some people here don't even use bayes and are doing much better. Look at your SURBL and RBL hits. These seem to be the heavy hitters for me. Take a good look at your headers and insure you're getting hits on these. If you're not sure, doing a search in the group for spamassassin --lint and -d and use those to see what's really going on. Once you get your scores up a bit drop me a line (off list if you want) and I'll get you everything I have on bayes training with setups like ours. HTH! Ken Ken Goods Network Administrator AIA Insurance, Inc. !****************************************! ! * Press any key to continue...NO, NO, ! ! NO, NOT THAT ONE!!!! !
