[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: possible mass bug filing: spamassassin 3

martin f krafft wrote:

>also sprach Adam Majer <adamm@galacticasoftware.com> [2004.10.06.1934 +0200]:
>>After running for a little while,
>>I wouldn't say it uses half a gig of ram. Something else is going
>I had -m10 passed to spamd for 2.64. When I upgraded, I left that in
>place. I almost hosed a server that went up to load 30 immediately.
>It took me 40 minutes to get a shell, another 30 to halt postfix and
>unload SA.
>Now it's running at -m5 (the suggested value) and I have not had
>problems since. Of course, now I only get half the throughput, and
>my queue has not emptied for a whole day because SA is unable to
>keep up.
Spamassassin 2.63 used less ram,

14615 spamd     16  15 23204  11M  4708 S N   0.0  2.3   0:24 spamd

so that would be about 40% of what spamassassin 3.0 seems to be using at
the start and about 1/4 of what 3.0 used in my little run below.

3.0 is also seems slower than 2.63 probably because new/better tests
were added..

I've also run spamassassin (spamassassin --mbox < mbox > obox), where
mbox is about 850 messages. It took spamassassin,

real    2m45.994s
user    2m34.211s
sys     0m1.949s

It also used a max of 40MB during the run.

This means that spamassassin 3.0 scans at a rate of about 5 message per
second on Athlon 2000+. Max thoughput for the day should be at about
400k messages, but probably <100k should be a better target.

Concurrency is not going to help you unless CPU is idle. In my case, -m
2 would use 100% CPU (not using razor).

- Adam

PS. 3.0 seems *much* better at detecting obfuscated spam. Where 2.63
gave an email 0.5, the new version gives it 7.3. Spam beware! :) On the
other hand, this is a type of spam arms race. Running spamassassin will
become more expensive as it improves.

Building your applications one byte at a time

Reply to: