trey valenta writes: > On Fri, Sep 05, 2003 at 04:56:45PM +0200, Jochen Tuchbreiter wrote: > > Did any of you analyze (profile) where most of the CPU in spamassassin > > is spent? > > I ran "spamassassin --lint" through Perl's profiler back in July. I know > this isn't the same as checking mail for spam/ham ...
BTW I would suggest instead profiling spamd -- "--lint" is not the case you want to optimize for ;)
Speaking of profiling, is there anything in the works in the future for some kind of built in profiling system ?
What I'm thinking of here, is something that can be turned on (in user prefs and system prefs) that adds a new header something like X-Spam-Performance, that lists at least (or hopefully :)
Total wallclock time
Total system time
Regex processing time
Bayes processing time (scoring a message, not learning)
Bayes learning time (autolearn)
DNS RBL times (wallclock time)
Seperate Razor/DCC/Pyzor wallclock times.
Also perhaps, which RBL checks timed out (if any) and whether Razor/DCC/Pyzor timed out.
I wonder how doable that is in the 2.6x series or is that a 2.70 thing ? (If at all)
At the moment the first two items can be gleaned from manual timing of spamassassin and spamc, but none of the others easily can, and also in environments like spamass-milter or amavisd etc the performance of spamc or spamassassin are somewhat erroneous anyway.
Essentially this would be off-by-default extra information/debugging information.
Regards, Simon
------------------------------------------------------- This sf.net email is sponsored by:ThinkGeek Welcome to geek heaven. http://thinkgeek.com/sf _______________________________________________ Spamassassin-talk mailing list [EMAIL PROTECTED] https://lists.sourceforge.net/lists/listinfo/spamassassin-talk
