[Mimedefang] Patch to Add Spamd Support

David F. Skoll dfs at roaringpenguin.com
Tue Jun 17 09:31:01 EDT 2003


On Mon, 16 Jun 2003, Richard Laager wrote:

> Attached is a patch which adds spamd support to MIMEDefang. Spamd is a
> daemonized version of SpamAssassin. The documentation is also updated by
> this patch to reflect the code changes.

Thanks.

> The main advantage to using spamd vs. the strictly Perl version of
> SpamAssassin (the default of MIMEDefang) is that one copy of SpamAssassin
> per machine can be shared. This eliminates the extra overhead of loading
> one copy of SpamAssassin per message that comes in (which is how
> spam_assassin_check) operates.

That's not correct; spam_assassin_check uses one copy of SpamAssassin
per slave, not per message.

> With the volume of mail our servers handle, we were finding that the
> startup CPU usage of compiling SpamAssassin on a per-slave basis was so
> high that unless we setup the machine to start a fixed number of slaves and
> keep them running, a sudden blast of spam would create a chain reaction
> that would bring the server to its knees and cause all of the messages to
> timeout (and be tempfailed).

Well, then you should start a fixed number of slaves and keep them
running :-).  Also, you can increase the maximum number of
requests/slave to a high number like 5000 so you don't pay the startup
penalty very often.

I'll look over the patch and consider it.  But I need hard numbers showing
a performance improvement, because I doubt it really helps.  It introduces
yet another socket over which the message has to be transmitted.  Although
it may save some memory during idle time, I need convincing that it
improves performance.

Regards,

David.



More information about the MIMEDefang mailing list