On Mon, 24 Mar 2003, Vernon Schryver wrote:
From: Matt Sergeant <msergeant(_at_)startechgroup(_dot_)co(_dot_)uk>
We average about 1 second per email. We process over 10 million emails a
day for about 1.2 million users. IIRC our average processor is a 1.2Ghz
Pentium.
Yes, this is an inordinately long period of time.
It certainly is! Are you user you're not counting other costs in that
1 second? For example, the 3 or more DNS lookups per incoming mail message
(not counting DNS blacklists) can be time consuming.
Right. We don't count those as they vary so wildly as to sometimes skew
the stats, and also we do those lookups in the smtp server, whereas
scanning takes place in the "backend" (this is qmail-ish). The DNS lookups
probably push it up a bit, though we try and keep large local caches where
possible.
I think the most interesting number you've offered is the 10
msgs/user/day average. What is the average spam load? For U.S.
retail ISPs it seems to be 70-85%. Are your numbers similar to
what I can see of other European ISPs?
We're an international company, so I can give you a fairly good worldwide
view...
Spam to overall ratio for December was 1 in 2.13 emails for the US. That's
detected spam (we detect about 95% with low enough FP rates to make errors
insignificant).
For the UK December was 1 in 6.0 emails. Unfortunately this figure gets
badly skewed because we generate about 1.2 million emails a day in
server-to-server notices (I've whined about why this is a bad idea, so it
may get changed eventually), and they end up in the UK pile.
For our Asian customers it was 1 in 6.9, though I'm not convinced by that
figure - the asian customers may have turned off our heuristics as they're
very aggressive against asian emails (or at least they have been in the
past).
Overall that gave us a December figure of 1 in 4.2 emails detected as
spam.
I can try and get more up to date figures tomorrow - these are just the
spreadsheets I had to hand. And I have a feeling these spreadsheets count
total emails as inbound and outbound, and since our customers don't spam
that skews things a bit too (or does it? - perhaps you'd like both figures
for completeness sake).
The only issue is generating these figures is time consuming for our stats
guy, and he is in demand a lot from marketing, so I have to subpeona (sp?)
his time ;-)
Matt.
_______________________________________________
Asrg mailing list
Asrg(_at_)ietf(_dot_)org
https://www1.ietf.org/mailman/listinfo/asrg