performance (was Re: file locking)

[Jeremy Blosser]

On Sep 18, Adrian Otto [aotto at aotto.com] wrote:
> I think that using a client/server model is perfect for this. In it's
> current stage of maturity, I think it's best for bogofilter to remain as
> simple as possible. We should put energy into developing the client/server
> functionality when we have users that start to suffer from performance
> problems.

What kind of mail loads are you currently getting feedback from?  My shop
is very interested in tracking the various fast bayesian implementations as
an alternative to our current anti-spam measures which are creating serious
load issues for us.  We see around 130k-150k messages/week, with 20+% being
spam.

For our part we don't care a lot about bogofilter having this or that
dependency as long as it's as fast and efficient as possible (as long as
all the dependencies are still Libre Software of course).

Anyway, there are of course other shops with much higher loads than us, and
maybe you're already hearing from them.  But if we can provide any useful
benchmarks or other data just let us know.  Currently we're just doing
basic stress testing via running every message through bogofilter -v and
bogofilter -s -v to see how big the spam database gets and how well
bogofilter and the server handle it.

For summay digest subscription: bogofilter-digest-subscribe at aotto.com