mailbox classificataion

Gyepi SAM gyepi at praxis-sw.com
Fri Jan 31 18:44:05 CET 2003


On Fri, Jan 31, 2003 at 10:23:45AM -0600, Fred Yankowski wrote:
> On Fri, Jan 31, 2003 at 04:29:23AM +0100, Matthias Andree wrote:
> > I don't believe that autodaemon will improve the performance when
> > registering big mailboxes, because we'll copy data around, feed it
> > through pipes and sockets which may again imply copying,
> 
> I would predict that those potential costs are tiny compared to the
> benefit of not repeatedly loading/storing the database for each
> message.

I would disagree. Keep in mind that the autodaemon code was started when bogofilter
was still using Judy and therefore paid a high cost for loading/writing the database.
As it turned out, ESR gave up on the idea just about when I gave him my db3 patches.

Remember also that BDB is quite fast. We pay the cost of database initialization
each time we open a database, but since the databases are not loaded into memory at
initialization, we pay nothing there. Locking obviously slows things down, but that's
to be expected.

Now consider that a daemon would still have to do everything that bogofilter does now
in terms of reading and writing to the database and on top of that we add the copying of data
back and forth. It would definitely be slower all around.

Having said all of that. A bogofilter daemon that only handled classifications
might be quite fast, especially if we use a fast readonly database like DJ Bernstein's CDB.

-Gyepi 





More information about the Bogofilter mailing list