OK... Thank you very match for all your time.
I am afraid the problem is in ldapserver. i dont know why stil, but i use another server and with the same stress the preprocessing never was up than 2 or 3 mails. delivering 10000 messages to local queue.
the openldap server are the same version, only slackware version change but i think this question is for another mail-list.
For this stress deliveris arount 20 or 30 mails for second what will be the best concurrencylocal and remote values?
i have put 500, but is too much slowly. and 40 too.
Thanks to the list/Claudio.¡¡¡¡¡
When i will get to put this system in production, i send my success with the all history to the list.
César García. Dept. Sistemas, IdecNet S.A. Centro de Gestión de Red. Edificio IdecNet. C/Juan XXIII 44. E-35004, Las Palmas de Gran Canaria, Islas Canarias - España. Tfn: +34 828 111 000 Ext: 340
Claudio Jeker wrote:
On Mon, Jun 14, 2004 at 11:31:50AM +0100, Cesar Garcia wrote:
I cannot see any problem in logs or in qmail-remote/local deliveries, the really problem is that although i can see x/500 in local and remote deliveries the queue/preprocceced is 30000/29800 more or less.
I am stressing de server in not real scenary and i can see that with a queue of 27000 injected emails, the queue preprocced at 600 email in one minute, or 10 email in one second.
a poor result ? i think yes, i put my ldap db with loggin 0 in memmory to test ldap benchmark, but no result...
the result is too poor.
... and not normal. My desktop PC with a crapy IDE subsystem did 10000 mail deliveries in less than 5 mins and the preprocessing took IIRC half of the time. See: http://www.nrg4u.com/qmail/qmail_ext_todo.png
this is a top of the server in stress situation.
11:30:25 up 1 day, 21:10, 2 users, load average: 8.49, 6.96, 24.35 133 processes: 126 sleeping, 7 running, 0 zombie, 0 stopped CPU0 states: 3.3% user 9.2% system 0.0% nice 54.4% iowait 31.4% idle CPU1 states: 3.0% user 10.4% system 0.0% nice 41.0% iowait 44.3% idle Mem: 2077224k av, 1554276k used, 522948k free, 0k shrd, 464240k buff 407344k active, 735740k inactive Swap: 3622636k av, 0k used, 3622636k free 654496k cached
OK, 54% and 41% in iowait tells me that your bottleneck is the disk. What OS, filesystem are you using and how is it mounted. You probably need to seperate disks, use better a better raid and/or a adapter with nv-ram cache. qmail-todo is hammering on the disk in a extrem fashion, a lot of open->read->close calls and fsync()'s.
As you see qmail does not need super high speed CPUs but a fast IO subsystem and enough ram to run fast.
