jdow wrote:
From: "Steven Stern" <[EMAIL PROTECTED]>

Please keep replies on the list.

In my former job, we ran 50,000 messages/day through sendmail on a sparc 20. It chugged a little, but it handled it. I think a decent Xeon box with a decent amount of memory could easily handle 50,000 messages plus the milters. In my current position, we've used greylisting in a qmail environment using some qmail addin that I can't recall right now. We dropped greylisting because our users were willing to trade more spam for zero delay in receiving messages from new senders.


As a scaling factor scanning one message via spamc on a 2GHz Athlon with
1G of memory took 2.77 seconds elapsed time. Interestingly two at the
same time takes about the same time. Three spam scans, one on an incoming
message, seemed to take about 3 seconds. I'd conclude MY setup runs at
about 1 second per message. It'd chug heavily at a 50,000 messages a day
load without uniform density throughout that day. I use 40 of the SARE
rulesets and per user Bayes.


Turn off network tests and look at those timings. SA doesn't take time because it's thinking about stuff. It's waiting for external servers to reply to its queries.



--

Steve

Reply via email to