On Wed, Oct 23, 2002 at 08:39:58AM -0500, Dallas Engelken wrote: > > On Wed, Oct 23, 2002 at 05:35:20AM +0200, > > [EMAIL PROTECTED] wrote: > > > > > hi list > > > > > > however, once the mails received go higher than 50 at one time, > > > the CPU load shoots up to 90% or higher. and when this happens the > > > mail server becomes very slow and sometimes users complain that > > > they are not able to download their mails anymore. > > > > > > i noticed that the CPU used by spamd when checking one mail can go > > > from 3 to 6 %. is there a tuning we can tune spamassassin so that > > > it doesnt burden the CPU too much? > > > > Seperate the two functions. Run spamd on a dedicated box with lots > > of CPU and memory. Disk is not terribly important. Just add -d > > spamd.box.domain to you spamc invocation on the mail server. > > > > Of course, you probably need to use a userprefs database in that > > case, rather than the userprefs in each user's home directory. > > It is working splendidly for my ISP in that configuration. We > > haven't seen any appreciable load increase on the mail servers since > > beginning use of SpamAssassin site wide. > > > Is spamd threaded? I see the PID jump around alot on high volume. > 44,000 out of 600,000 messages were not scanned via spamassassin... I > assume because the spamd daemon did not answer. I'm running it via > daemontools, but have tried it both ways.
If you use the -m parameter to spamd it has a tendency to die. DaemonTools restarts it but a few slip through while spamd is dead. I see that effect on my system. When I ran it without the -m option, (and without much RAM), a spam run on my 2000 users could send the spamassassin box so far into swap that it took it 8+ hours to recover. I haven't quantified how many messages go unfilterred yet. I get 5 to 8 per day in my mailbox which recieves mail for 10 or so role accounts. I have not tried removing the -m option since upgrading the RAM to 1GB. I have also disabled the razor checks because they were adding a lot of latency. My average message processing time with razor/razor2 was 6 seconds, without it is 1.1 to 1.5 seconds. The default required_hits for my users is 8. A handful of individual accounts have a required_hits of 5. I'm running SpamAssassin 2.41, so I'm probably letting a significant number of spams through that 2.43 would catch. I just don't get a feeling of trust for 2.42 and higher in the false positives category yet. I'm a late adoptor by nature. Here is a typical summary for a weekday: SpamAssassin statistics for 2002/10/22: Time spent processing all messages: 70424 Total messages processed: 44517 Average processing time per message: 1.58196 Average message size in bytes: 9261.85 Everyone has had 20257 non-spam messages and 24260 spam messages. 54.496% spam messages. Here is a typical weekend report: SpamAssassin statistics for 2002/10/20: Time spent processing all messages: 32233 Total messages processed: 28454 Average processing time per message: 1.13281 Average message size in bytes: 7413.78 Everyone has had 8841 non-spam messages and 19613 spam messages. 68.9288% spam messages. -- Scott Lambert KC5MLE Unix SysAdmin [EMAIL PROTECTED] ------------------------------------------------------- This sf.net email is sponsored by: Influence the future of Java(TM) technology. Join the Java Community Process(SM) (JCP(SM)) program now. http://ads.sourceforge.net/cgi-bin/redirect.pl?sunm0002en _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
