Hi to all exim admins...
I've setup exim4 on Debian (2.6.20-15-generic, x86_64) box what will (and
is) accepting large amount of SMTP connections - in peaks 300+ per second,
100/s on average. 99% of mail will be only few KB's in size.
Exim uses Maildir's that are located on ramdisk, mainlog/rejectlog is
symlink'ed to /dev/null and "log_selector = -all" is also set in configs.
"smtp_accept_max" is set to 1000, "smtp_load_reserve" is rised to 100.
"rfc1413_query_timeout" is set to "0s". Don't know other ways to optimize
it...
There're no others services on the box, "top" shows no processes using more
then 0% CPU. us/sy/ni/wa/hi/si/st stats are all low, CPU idle is almost
100%... and it doesn't metter is exim4 running or not! The only thing
changes when i do "/etc/init.d/exim4 start" - top's load average goes to
50+, usually ~100 in few seconds. "ps -e|grep exim|wc -l" shows 50-300
(runnig exim4 processes). System remains stable and responsible as with load
avg <1 (before firing exim4).
Hardware inside: Intel CoreDuo2 E4300 (1.8GHz), 4GB of RAM, SATA disks (but
that doesn't metter, i've Maildir's on ramdrive), Intel PRO/1000 NIC. Tried
to optimize IPv4 some vars thru sysctl, nothing helps.
Digging thru maillist for few hours... the only thing somehow related to my
"problem" was the post that says "it's ok with that load avg because of
large exim4 processes/concurrent connection". Post was from 2004 year...
I can leave the system as is... but want to hear from people who are hosting
LARGE SMTP relays using exim4, do they have similar situation with "load
average".
Thanks!
BR,
jamez.
