What kind of servers do I

...

need to run SpamAssassin for my large site

...

There are so many different variables, it's very difficult to make suggestions. Some factors that will affect the answer are:

1. How many users are there?
   2. Will they all be having their mail scanned?
   3. How much mail does each user receive daily (number of messages, total size in bytes)?
   4. Does mail arrive evenly throughout the day, or is it "bursty"?
   5. Are you doing other processing on that server (POP3/IMAP mailboxes, virus scanning, ...)?
   

?

Obviously, every site is different. Here are some details on one large scale deployment that you may find useful as a benchmark.

MWEB is the largest ISP in South Africa, with approximately 350,000 mailboxes, all of which get filtered through SpamAssassin.

Their setup entails 8 servers running Postfix, 12 servers running amavisd-new, and an additional 5 servers running Trend Micro and Vexira anti-virus. Their 25 servers are Dell Power Edge 2650's with dual 3.0 GHz Xeons, 4 gigs of ram, and 5 SCSI drives in a RAID 10 configuration (1 hotspare). They're running a custom built Fedora Core 2, 2.6.7 kernel .

Here are their statistics for April:

Year  Month Messages Blocked  Messages Passed  Total Messages  Percentage Blocked 
2004   Apr     106,710,305     25,065,604       131,775,909         80.98 % 

Thus, their long term average is 131,775,909 / 30 / 24 / 60 / 60 = 50 messages per second. Their peak rate is up to about 300 messages a second, or 6x their average.

Note that while amavisd calls Mail::SpamAssassin directly, it runs Perl as a daemon and should offer performance quite close to spamd. This suggests that a single brawny server running spamd can comfortably handle 5 messages a second and may peak up to 30 messages per second.

More info

Please contribute details on your large installation. It's fine to do so anonymously if necessary.

Contributors

• DanKohn interviewing camis at mweb.co.za

...