Re[2]: Using SA over network info

[Robert Menschel <Ro...@Menschel.net>]

Hello Gavin,

Wednesday, October 27, 2004, 9:33:07 PM, you wrote:

GC> Hi Robert,

>> Why couldn't you simply install another perfectly running copy of
>> SA on server A, and let the users on that machine call spamc
>> directly on that machine?  

GC> I'd prefer the one copy so I can maintain the one ruleset,
GC> Bayesian database, AWL etc.

Understandable.

I manage the emails for three domains, and have the same ruleset
configuration for all three. When I make a change, I validate it on my
PC, upload it to my family domain for a second-level validation, and
then upload it to the two other domains.

There's some debate concerning whether it's better to have one global
Bayes database or multiple independent databases. Each person's spam
is different. IMO the big question is really how you'll be feeding the
Bayes database.

I would think that on your server A (multiple users), it might be best
to allow each user to have his own Bayes database. When Bayes works,
the databases are reasonably small (my largest Bayes database is 26
Meg), and this allows for much flexibility. It shouldn't be too hard
to send copies of your sa-learn files from system B to system A and
apply that to each active database in turn, if you're interested in
that approach.

I'm also thinking that AWL is probably best handled on server A by the
individual. User A1 may want email from printpal, while user A2
considers printpal to be a horrible spammer. Separate AWL databases
will allow them to maintain consistency within their own email paths.

The questions all probably boil down to resources: do you have the
machine and/or personnel resources to support two SA servers, and to
transfer data between them as appropriate for your environments?

To me it seems that running SA independently on both machines,
occasionally sharing sa-learn data, and sharing rule set files, should
be a workable solution, and in my systems trying to serve both types
of environments with one SA server would be problematic. But there may
be other attributes in your systems which reverse that consideration.

I know what works for me, and like with almost everything else when
dealing with computers and software, YMMV.

Bob Menschel


GC> Cheers

GC> Gav







-- 
Best regards,
 Robert                            mailto:Robert@Menschel.net