In message <m0vXt85-000GXNC@cichlid.com>, Andrew Burgess writes:
>This is not necessarily 'nothing'. I don't know how bulk_mailer works
>but if it runs as one process with internal threads (as opposed to
>forking itself for each job) then a load average of 0.7 could mean
>70% CPU usage if bulk_mailer is all that is running.
The load average has absolutely no correlation with CPU usage.
>Load average is how many processes are ready to run. A runaway CPU
>bound process on an other wise idle system will give a load average
>of 1.0 but be consuming 100% of the CPU.
You can get a high load average with a CPU usage of 5% but heavy disk
or paging activity.
You can't use load or CPU as a list performance metric. The main things
that really count as a metric here are 1) average latency (time to
deliver) and 2) message thoughput (deliveries/hour).
--Dave
References:
|
|
