I tried doing this on my Spamassassin 3.0.1, Postfix2, Redhat Enterprise box on dual 1.5ghz xeon 1gb memory hardware, and i'm not sure if it made any difference.
I;ve been quite disappointed with spamd performance on a box with that much grunt (?). During peak times we get hit with about 1msg/sec and spamd cant quite keep up with that, with the incoming mailqueue growing to 1000+ a times. Spamd logs report spam identification taking from as little as 0.2secs to as long as 5.0secs, with the average being 1.8 seconds or so. Is this the type of performance i should be expecting? The above stats are for spamd running in default config, without playing with max-children or max-connections. I also have Anomy sanitizer (a filter that removes dangerous email attachments) (no damonized version of this available) along with spamd which get launched form the same script, so the max-children setting would also apply to Anomy. It also chews some cpu, but not much because no textual analysis is actually needed for it. Disabling it doesnt seem to have an impact on the box's performance. Anyone out there have any tips on how to tune Spamassassin for better performance? thanks in advance, Regards, Dimitry -----Original Message----- From: Paul Crittenden [mailto:[EMAIL PROTECTED] Sent: Wednesday, 17 November 2004 7:29 AM To: users@spamassassin.apache.org Subject: Fwd: Re: Tuning SpamAssassin Well, I'm not sure exactly what fixed my problem but it is working fine now. I'll just need to pay attention and see if there is any other tuning that needs to be done. I increased the number of children to 35, changed the max-conn-per-child to 1, removed the lock for spamassassin and added DROPPRIV to my procmail script. One of those 4 fixed my problem or perhaps all of them. Now emails go into and out of the queue pretty much like before. Thanks for all of your help. >At 08:37 AM 11/16/2004 -0600, Paul Crittenden wrote: >>I'm running SpamAssassin 3.0.1 on a Compaq Alpha running Tru64-Unix, >>sendmail 8.12.10. are there any guidelines to figuring out how many >>children you need when running spamc/spamd. I know that having to many is >>as bad as having to few. I have tried 10, 25 and 30. Thing seem to be >>going better as I continue to increase the number of children but how do >>I know when I have gone to high and need to decrease the number? > >Check your swapfile stats. (use top or tool of your choice) If you're >digging into the swap to any significant degree, you've gone too far. > >Note: many OS's will page out stuff that's not been used in a long time. >Compare swap usage with your "available" ram that can be used if programs >need it (free physical ram+buffer ram). There should be more available ram >than swap used. > >Basicaly you can keep running more spamd's as long as you're not running >out of ram. Once you run out of ram, it starts swapping, and things >quickly grind to a halt. Leave some extra megs free, at least 40mb if you >include buffers, that way bumps in memory load won't slow you down.
