How to determine what MaxDaemonChildren value to use
Amadej Bukorovic
At my company we`re running a web server that also serves e-mail postcards
and last night I noticed that sendmail "shuted down" because of high
traffic.
sendmail[70]: rejecting connections on daemon MTA: load average: 12
This is what I found in my logs and after examining them and the web server
statistics I determined that the system could handle a bigger load average
than 12. Naturally I should tweak sendmail so it won`t reject connections at
such a low load average (I think?), but I am a bit puzzled about how to
determine at what system load average the server will still function
normally and at what load will it slow down.
From my understanding the system load average of 4 could for instance
totally lag a, let`s say, Intel 486, meanwhile a system load average of 30
could work fine on a Pentium III 600... that is if I understand the meaning
of the system load average correctly from what I have read(please correct me
if I`m wrong).
So could someone point me to some documentation or explain to me how can I
determine at what system load average should sendmail really start rejecting
connections on our server?
Thank you for the help in advance,
Amadej Bukorovic.
Neil W Rickert
>Hello everyone!
>At my company we`re running a web server that also serves e-mail postcards
>and last night I noticed that sendmail "shuted down" because of high
>traffic.
>sendmail[70]: rejecting connections on daemon MTA: load average: 12
If you have a variable email load, then configure sendmail
to use DeliveryMode=queue and have reasonably frequent queue runs
(say, every 5 minutes). This tends to even out the load, and avoid
sharp peaks. The MaxDaemonChildren control will be more effective
in that case, since there won't be lots of uncounted grandchildren
processes doing background delivery.
If you are generating mail with cgi scripts, that is probably not
being counted in MaxDaemonChildren.
>This is what I found in my logs and after examining them and the web server
>statistics I determined that the system could handle a bigger load average
>than 12. Naturally I should tweak sendmail so it won`t reject connections at
>such a low load average (I think?), but I am a bit puzzled about how to
>determine at what system load average the server will still function
>normally and at what load will it slow down.
Be careful with this kind of thinking. The load average is the queue
length. The longer the length of the queue of processes waiting for
CPU time, the slower everything will run. It depends what is in the
queue, as to whether this is a problem. You might be able to have a
load average of 100, due to lots of looping processes with a good
"nice" value, so that they don't cause problems with other activity.
A load average of 1.5 can be a big problem if it is due to a process
using gobs of memory and forcing the system to do heavy virtual
storage swapping. In other words, load average is not the whole
story.