Jakob Hirsch wrote: > Hi everyone, > > I've searched a little through the archive but found not much about that. > Right now I'm using DBM for bayes. I know the officially recommended > storage is a SQL DB (though I don't know why). I have a mysql server > running on my mail system and tried using it for bayes a while ago (with > SA 3.1.0) but didn't see lower scan times. For a local RDBMS I think > SQLite is a very good thing, but there was a list post where somebody > said ist not suitable for SA (though README.bayes mentions it). > According to the bayes benchmark in the wiki > (http://wiki.apache.org/spamassassin/BayesBenchmarkResults), SDBM is the > fastest store (though there are no sqlite numbers), but I wonder why > it's not mentioned much. I'll probably run the bayes benchmark myself > (with sqlite, sdbm and dbm), but somebody has probably some experience > to share.
I personally use sdbm on my production servers. I chose it for better speed than dbm, without the extra ram overhead of mysql. I posted some questions about this in December, with no real response: http://mail-archives.apache.org/mod_mbox/spamassassin-users/200512.mbox/[EMAIL PROTECTED] I'd be interested in hearing any results you get from using sqlite... As for SDBM, I'll warn you that if you're not using SA 3.1.1, there's a bug in sa-learn --restore that affects sdbm. http://issues.apache.org/SpamAssassin/show_bug.cgi?id=4670 That said, I've been running bayes with SDBM in production since December, using SA 3.1.0, and am fairly happy with it.
