[Mimedefang] Bigger MAX_QUEUE_SIZE, why is there a hardcodedlimit ?

[Stephen L Johnson]

On Wed, 2007-08-08 at 22:15 +0100, Paul Murphy wrote:
> Mack wrote:
> 
> > each mail (inc SA should take way less than 1s inc virus scanning).
> 
> Woah!  Sweeping generalisation alert!
> 
> If you have a monster server with 64Gb of RAM, a super-fast processor, a RAM
> disk for everything, and only one e-mail arriving at a time, you might just
> (and I really mean *might*) get the whole of SA down to 1 second, but to
> manage this for the whole message is completely impossible in my experience.
> 
> Does your Spamassassin implementation actually do any rule processing?  DNS
> RBL lookups?  Bayesian analysis?  Auto whitelisting?  I can't believe it
> does.
> 
> Add in a message size dependency for both SA and AV, and you're waaaaay over
> 1 second.

Processing times can be down under a second. I have an incoming mail
cluster which mean scan times that do stay under 1 second for a
significant amounts of time. And scan times are generaly well under 2
cdonds. I have most of the milters running on a pair of somewhat hefty
back-end servers (Dual 1.66GHz Power 5 P-series). The front-end servers
just handle the mail tasks. 

I found it too hard to balance the 2 types of loads on the same box
(sendmail and mimedefang). So I split the mail handling from the heavy
lifting. It was easier to tune the systems for the load they'll carry.
And as a result I found I could use some lower end servers on the
front-end for mail handling. 
-- 
Stephen L Johnson  <stephen.johnson at arkansas.gov>
Unix Systems Administrator / DNS Hostmaster
Department of Information Systems
State of Arkansas
501-682-4339