[Mimedefang] compare mimedefang to mailscanner
John Rudd
john at rudd.cc
Tue Jan 16 13:09:45 EST 2007
Kenneth Irving wrote:
> I've found Sendmail+MIMEDefang+clamd+spamd a very efficient combination.
> Tried Postfix+MailScanner+clamscan in another computer and it's very slow,
> because Mailscanner doesn't work as a deamon, and doesn't use Clam as a
> deamon either, but relies on running clamscan for every email. The overall
> result is extremely slow.
>
I forgot to respond to this part:
This is entirely what I was talking about with "economy of scale". With
MailScanner, if you're scanning messages in such small batches, and so
infrequently, that you're spending more time with the clamscan
start-up/initialization overhead, then it will seem like MailScanner is
slow.
On the otherhand, if you're scanning 100's of messages per minute, in
100 message batches, then the start-up overhead of clamscan or sweep is
so insignificant compared to the actual scanning cost, that you get a
HUGE benefit from this economy of scale. This actually far outweighs
the IO bottleneck of clamd's socket. MailScanner doesn't have any
support for clamd exactly because clamd would slow down MailScanner in a
BIG way.
(a few years back I tried to do a comparison between
MailScanner+clamav_perl_module and MIMEDefang+clamd to see if either
could be improved by the other ... and I couldn't -- each of them is
going in such a different IO model that you can't really imply something
about one's model based on your experiences with the other model... the
only answer is: MailScanner is fastest with the perl ClamAV module, and
2nd fastest with clamscan ... but clamd is a significant liability with
MailScanner; meanwhile, MIMEDefang is _exactly_ the opposite -- using
clamscan with MIMEDefang causes a significant slowdown because you pay
the start-up overhead of clamscan with every message ; using the
experience of one to try to speed up the other will just lead you to
headaches)
More information about the MIMEDefang
mailing list