> The problem with SpamAssassin as I understand it (and in my experience)
> is that it does a lot of stuff with each mail and gets bogged down very
> quickly. Dspam's main touted feature is its low overhead and ability to
> handle high volumes efficiently (I have no idea how well it lives up to
> that, which is one reason I want to try it). Tests I've done with SA
> indicate that it can easily take 10+seconds to process one email.

I've been using dspam in production for about 3-4 years now, maybe a bit 
longer, and I can say I'd never go back. In that time, there was a stretch 
of about 2-3 years where not a single spam made it through to any users' 
mailboxes. Now the spams are getting a bit trickier to catch, so we see 
maybe 1-2 spams slip through _per month_ across all users, but its also 
stopping about 3000-4000 incoming spams per-day. I also added graymilter 
in front of that, and it dropped the incoming spam by about 90%. The 
combination of the two has made a HUGE difference. 

The speed of dspam is imperceptable. The main benefits (for my needs), is 
that it allows the user to manage, train/retrain and condition their own 
personal spam thresholds. If one user WANTS all of that HTML, 
web-bug-laden email, they can have it. If another one wants none of that 
garbage, they can have that too. The web-based queue lets them tweak it 
however they choose. 

If a spam slips through, they just forward it to 
'[EMAIL PROTECTED]' for example, and dspam will retrain it so they 
no longer receive emails of that type any longer. If they made a mistake, 
they can go to the web-based quarantine or queue and resend it back to 
their inbox (this also applies for false-positives). dspam will retrain it 
with a lower score and let it through. It also has automagic whitelists, 
greylists and other features. 

Back when I ran SA, we were using 15 RBLs, about 40 custom rulesets in 
Sendmail, and SA was trained down to a 2.0 threshold. After 1-2 years of 
using it, it was still only scoring 85%-90% accuracy, and we were getting 
20-30 incoming spams slip through every day. 

After replacing SA with dspam and training it with the spam/ham corpus, we 
were at 98%+ within a week. It's steadily gotten better since. I think 
we're at 99.835% accuracy now (based on running dspam_stats a few moments 
ago).  We're also using no RBLs, no custom rulesets, and I no longer have 
to shroud my email address with standards-compliant "tricks" to stop 
spammers from harvesting it. 

I can put my email address anywhere and everywhere, and not worry about 
spam... because I simply don't get any. 

I *HIGHLY* recommend it, if you're on the fence. jonz (the main author of 
dspam) in #dspam on IRC is a great guy, and has been very helpful with all 
of my questions over the years of running it. 


David A. Desrosiers
Linux on Power Developer Program Manager
[EMAIL PROTECTED]


_______________________________________________
Peterboro mailing list
[email protected]
https://mailman.lug.org.uk/mailman/listinfo/peterboro

Reply via email to