Re: Running SpamAssassin on an old Pentium
on Sun, Apr 20, 2003 at 04:33:15PM -0700, Alvin Oga (aoga@Maggie.Linux-Consulting.com) wrote:
> On Sun, 20 Apr 2003, Craig Genner wrote:
> > Depends on what the 80% is of, 80 e-mails out of 100 isn't all that bad but
> > 800 out of 1000 would be to much. It also depends on respomse time, if you
> even 8 out of 10 is bad ... because the important info they needed
> was considered spam and dumped.... bad thing to do ..
Read the post, Alvin. Question wasn't "what's the false positive/false
negative rate", it's "what's the total mail load"?
My thoughts: P-133 is probably a bit on the feeble side, though for a
small community (up to several hundred recipients) it's likely
sufficient. Spamassassin is remotely accessible though, and you can
set up a spamd client on another host on the local network. Parsing
mail is the expensive part, and setting limits on the maximum size of
messages parsed is likely to help. Few spams are particularly large
(and in response to more effective spam blocking, many are now *very*
small, just a URL, essentially). Disabling network lookups may save you
some of that 36K dialup bandwidth. If you run your own domain,
SMTP-time mail rejection (set your limits high -- say 10 or better) may
be of interest. Configuration of this is MTA dependent, Exim 4.x can be
set up this way. Suggest you don't reject mail to abuse@ or postmaster@
your domain, regardless of spam signature.
I've run SA as spamd/spamc on a P-200 box with 15k recipients, 40-80k
messages daily. Main problem was tuning qmail to not spawn too many
simultaneous spamc clients. At worst (unoptimized) peak, we'd see loads
of 14+ and delivery delays of ~5-15 minutes. With tuning, this was
reduced to < 1 minute and peak of ~4-5 mid-days.
Karsten M. Self <firstname.lastname@example.org> http://kmself.home.netcom.com/
What Part of "Gestalt" don't you understand?
"Rob McKenna was a miserable bastard and he knew it"