Ed,
> I understand your comment on averages. I'll have to write the script
> in perl which will give me more functionality. I bet I can find some perl
> modules that can do some of the statistical leg work.
Brute force method:
sort an array and pick the middle element -> median
perl -e '@x = sort {$a<=>$b} (1,2,3,4,5,6,7,8,9,10,11);
printf("q10=%d, median=%d, q90=%d\n",
$x[$#x*10/100], $x[$#x*50/100], $x[$#x*90/100])'
q10=2, median=6, q90=10
> Here are timing averages from yesterdays log for 1 server.
> Timing Breakdown for 45975 Records
Thanks, very good.
> Timing Type Average Record Count
> -------------------------------------------------------
> SA-check 1530.5 ms 40168
As expected, SA is the largest resource hog,
anything saved here would be effort best spent.
> best_try_originator 881.6 ms 127
DNS lookup. Probably not worth worrying, as it is only wait time (no CPU),
so CPU can be used meanwhile by other processes. Still, having a good
nearby DNS caching resolver pays off for other tasks as well. It needn't be
on the same host, just withing the same LAN. Sometimes it is even better
to move it off the busy host.
> get-file-type105 740.5 ms 2
> get-file-type111 343.2 ms 8
> get-file-type100 178.0 ms 2
...
> get-file-type3 18.4 ms 2958
> get-file-type2 17.2 ms 16652
> get-file-type1 16.5 ms 24896
Just fine, time spent by file(1) utility is proportional by the
number of files to check, and is quite reasonable.
> SMTP-DATA 155.5 ms 45975
That's about as fast as it goes. The amavisd-new-2.3.3
brought about 35% speedup in this section, but you already
have that.
> fwd-data-end 97.6 ms 45958
> fwd-connect 14.2 ms 45959
> fwd-mail-from 2.6 ms 45959
> fwd-data 2.3 ms 45958
> fwd-rcpt-to 2.0 ms 45959
> fwd-rundown 1.5 ms 45958
Very well, your Postfix on port 10025 is responsive and fast.
> AV-scan-1 15.8 ms 41758
Very good. Clamd I presume.
> mime_decode 13.9 ms 45975
> decompose_part 2.8 ms 38020
Quite reasonable.
> main_log_entry 12.5 ms 45958
Just fine, as expected.
The rest is negligible compared to the above.
Concentrate on SA - but there is probably not much to be done
there after you already stripped it down to bare essentials.
Its CPU usage dominates your load.
Adding some network tests and Bayes on SQL can improve the quality
of results significantly. Network latency is not a bad thing, it can be easily
be compensated for by increasing the number of child processes.
Mark
-------------------------------------------------------
This SF.net email is sponsored by: Splunk Inc. Do you grep through log files
for problems? Stop! Download the new AJAX search engine that makes
searching your log files as easy as surfing the web. DOWNLOAD SPLUNK!
http://sel.as-us.falkag.net/sel?cmd=lnk&kid=103432&bid=230486&dat=121642
_______________________________________________
AMaViS-user mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/amavis-user
AMaViS-FAQ:http://www.amavis.org/amavis-faq.php3
AMaViS-HowTos:http://www.amavis.org/howto/
