[aklug] Re: Reasonable average server load?

On Wed, 23 Jun 2010, Royce Williams wrote:

> Two caveats:
>
> 1. Multiple processors or cores can yield different non-blocking loads.
>
> For example, on some OSes, pegging four CPUs using only four processes
> will yield a load of 4.0 on some OSes (with nothing blocking).

Which is why I run SAR everywhere. The load-avg lines in the SAR report
also report the process list size as well as the run queue size. If the the
run queue average is 1 or more for the day you know you've a capacity
problem somewhere.

Either way, UNIX systems aren't mainframes, in non-blocking situations where
CPU utilization is the primary component of the load average value the
target for a multi-tasking work load should be about 20%. Going over that
is going to virtually guarantee latencies when your process list size is
churning.

All of which, though, goes to show you how variable this can be, because no
metric works for everything. If you have a dedicated server running a
single application that's limited to a specific number of threads or
processes keeping the system pegged at 4.0 isn't a big deal. In fact,
that's probably optimal for getting the maximum productivity for the
hardware. It will be, however, for a general purpose system running
multiple applications.

> 2. If this is a virtualized environment, the activity visible to you
> (even as root) may only represent a small fraction of the actual
> processes and I/O running on the system.

Which is why running SAR on the host system is much more important.

         --Arthur Corliss
           Live Free or Die
---------
To unsubscribe, send email to <aklug-request@aklug.org>
with 'unsubscribe' in the message body.
Received on Wed Jun 23 12:10:17 2010