[EMAIL PROTECTED] wrote on 12/07/2005 13:53:28:

> Jim Maul wrote:
>
> > I dont see anything over 4.0 seconds there with the average around 2.0
> > seconds maybe?  This is with extra SARE rules, razor, and URIBL
> > lookups.  This is running on a p4 2.8ghz with 512mb ram.
>
>
> Yup - everyone likes to say "SA is slow" without understanding why. It's
> nothing to do with it being in perl - it's everything to do with all the
> best (well, most) pieces of SA involving NETWORK TESTS. All those
> DNS-based RBL, SURBLs, etc lookups.
>
> We've got spamd servers running in Europe, USA and here in NZ - all on
> "fast" hardware. And yet why do you think SA returns in < 2sec in the
> USA and yet takes 4-10sec in NZ? Yup - the fact that NZ is the Internet
> equivalent of the Moon :-( Latency is a real killer for the network
> tests. If I cared, I'd try to bring in a direct DNS feed for the RBLS
> being used  - but basically all of them would need to be local to see
> the performance benefit - and SA uses a lot these days. Adding 5+ sec
> per message isn't killing us - so I'm not bothering. And of course I
> could always disable all the network tests - but that effectively makes
> SA useless - so I can't imagine doing that.
>
> SA is "slow" because it does a lot. Some commercial antispam products
> are a lot faster than SA because they don't have "network tests" - their
> equivalent of RBLs/etc are all done via feeds - so are instantaneous.
> Highly inefficient - but instantaneous.

One cool alternative I have been using to avoid SA network tests is use a solid tcp.smtp content.
After investigating how I could take good advantage of this, I get known a sf.net project called spfilter (www.sf.net/projects/spfilter).

Instead of harvesting into all its file, check out contents from http://spfilter.sourceforge.net/data/output/ to qmail tcp.smtp (DEFAULT.qmail_uce.bz2). Notice that this 1.8M file will expand to a file over 100M.

I can provide more detailed information if someone wants to use it painless.

Regards.

Reply via email to