[Mimedefang] replicating bayes db?

[Ole Craig]

Hi -
	I'm having a bit of difficulty with SA's bayesian filter as
called systemwide by MD, and I'm hoping someone can help me figure it
out.

	I've just set up a second MD host, which will soon become the
main SMTP gateway for our network. I'd like to copy the bayes database
from my current mailserver to this new one, because the current bayes
DB is quite accurate. Both servers are running SA 2.63; the new
machine is running MD 2.42 (because that's the most recent pre-built
package available) and the old one is running MD 2.43 from source.
I've tried just copying the database files from the old to the new,
which didn't seem to work, and I've tried copying them and then
running sa-learn --import --dbpath on them and using the resultant
bayes_* files. This seems to work, in that spam scores highly on the
BAYES tests, but there's a disparity between the scoring on the two
machines: if I take a recent spam message and run it through both
machines, on the old server it will trigger the BAYES_99 test, and on
the new server it will trigger the BAYES_90 test. Since there's a 2+
point difference between the two, for a brand-new spam that hasn't yet
hit the SPAMCOP URI list or been added to the razor or pyzor
clearinghouses this 2-point spread often makes a difference in terms
of crossing the spamtag threshold. 

	Any ideas as to why I'm not getting the same bayes score if
I'm using the same database? Is there a recommended procedure for
copying a bayes DB? I couldn't find one in a few passes at the
manpages..


		Ole
-- 
Ole Craig * UNIX, linux, SMTP-fu; news, web; SGI martyr * CS Computing
Facility, UMass * <www.cs.umass.edu/~olc/pgppubkey.txt> for public key

   Need a seasoned *NIX admin in the Denver/Boulder area? Hire me!