Re: System loading.

Page principale
Supprimer ce message
Répondre à ce message
Auteur: Nigel Metheringham
Date:  
À: John Horne
CC: exim-users
Sujet: Re: System loading.
JHorne@??? said:
} The system is (will be) our main mailhub...it does nothing else but
} send/receive mail. I checked through some of the exim code, and found
} references to kernel functions for system loading, but after that I
} got lost as to where to look. I also noticed that the CDE startup
} screen and BSD "w" command refer to system loading, but again could
} find out nothing else.

Ahh - you are asking at a different level to my previous answer...

load average is one of these magic numbers, and you will find definitions
for it along the lines of "the average number of jobs in the run queue".
This means that each time the kernel came to schedule another job, there
were on average <n> processes in a runnable state (ie not waiting on I/O
etc) at the time. <n> is then the load average, and this can be an
average over specific periods of time - normally 1, 5 and 15 minutes (the
value used by exim will be the 1 minute load average).

If you have a CPU bound process - no IO to speak of - then it will
generate a load average of one all on its own, since it will always be
runnable. An IO bound process is rarely runnable and so contributes
little or nothing to the load average.

exim processes are typically IO bound.

If the load average is one or more, then there tends to be little or no
CPU idle time, and in theory (but not necessarily in practice) adding CPU
would help things go faster. In practice you tend to let the load average
go much higher than one before you start thinking about adding CPU
(assuming lots of processes - one process thats a CPU hog can only ever
generate a LA of one, so if you like doing CPU intensive single process
stuff then my comments above are rubbish).

I would say that for an exim only box, either leave the LA stuff
untouched, or at the sort of levels I suggested in my last mail. Then
have some process which monitors the LA and see what it does in practice
over the day (the snmp stuff may be able to do this on Solaris, or you can
knock up a script - something like "uptime|logger" and run that from cron.
If you can get the box to tell you when the LA is high and then examine
it to see why, then you can slowly work out what you expect to see in your
situation.

[if you have a multi-CPU box then load averages etc get far more
interesting]

    Nigel.


-- 
[ Nigel.Metheringham@???   -  Systems Software Engineer ]
[ Tel : +44 113 251 6012                   Fax : +44 113 224 0003 ]
[            Friends don't let friends use sendmail!              ]




--
* This is sent by the exim-users mailing list.  To unsubscribe send a
    mail with subject "unsubscribe" to exim-users-request@???
* Exim information can be found at http://www.exim.org/