Carlos E. R. wrote:
> I do, and I did tell you why. I'll try again because you do not > understand. Actually I do understand, as I noted before. I have even run a system in the very way you described, using amavis only for virus, and using spamc/spamd for spam detection. I still say this is an odd corner case. I'm not saying it won't work. > Amavis uses one single bayes database for all the users in the system, > whereas spamc called from each user procmail uses an independently trained > database for each user. This type of setup can be individually tailored to > the type of spam received by each one. My system is small, so I can get > away with that. > > So, I have disabled spam scanning in amavis, it doesn't load that part. I > would like to know how to tell amavis-new to use a different bayes > database for each user, but I don't know how. So I can't dispense with > spamd, unfortunately: the bayes spam checking method is the one that makes > here the biggest contribution, so I need it. > > The mail setup at SuSE is of the amavis-new type you propose, with one > global bayes database. It doesn't work well, it has had to be disabled > for the lists because lots of mails were marked as spam and were not. That > proves that is not a good method. I'm currently supporting a number of sites that use suse mail servers with the single bayes database for all the users. The largest site has 13,000 users, and in practice there doesn't seem to be a problem with, as you say, one man's spam being another man's treasure. For the most part spam is spam. *everyone* thinks the viagra ads are spam. For the case where someone might want to get viagra adds, that is why the per-user whitelists exist; the individual user can exempt specific senders or domains from spam checking entirely, for messages sent to him. > Seems nice. What about the bayes database? > > I know that amavis can be independently configured for each user. But I > haven't read about doing the same with the bayes database. It's a single bayes database, but the individual whitelists and per-user thresholds help make that not a problem in practice. Joe -- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
