Greg A. Woods wrote:
> At Sun, 23 Jul 2006 20:17:24 +0100,
> Mark Cullen wrote:
>> Oh, I know :-) In my defense, the speed benchmarking is just for me 
>> really, and I am quite happy comparing rough numbers with a benchmark 
>> run several times for each different configuration, just to get a 
>> general feel of the difference in performance. If I were to be doing any 
>> serious benchmarking for the NetBSD folks, which might help them improve 
>> performance, then I would certainly do it properly! :-)
> 
> BTW, if you really want to get a good feel for how your system performs
> under various disk I/O loads then benchmarks/postmark is the test you
> should be running.  You can easily adjust its parameters to more
> realistically model the normal production loads your system(s) will
> experience.
> 
> (You might have to search for and manually download the source before
> you build the package.  I wish NetApp hadn't done whatever they've done
> to try to stifle its maintenance and distribution -- sure doesn't make
> me want to buy or even use any of their product any more!)

Cool. I will have a look for that, thanks.

> 
> 
>> By externally observable do you mean listening for hard disk activity 
>> and such?
> 
> No, I meant by not having to dig through kernel data structures with a
> debugger and watch how often various kernel threads execute.  :-)

Oh right.

> 
> 
> As for your system's ~1.0 load average, well the only thing using any
> significant amount of cpu, besides smbd, is mysqld, though I can't see
> it keeping the load average up that high and yet still only use that few
> minutes of CPU time in two days.  However if you can run without the database
> for a while, see if killing it drops your load average down to "normal".
> 

Already tried killing off mysql for a while. It appeared to make no 
difference at all, just like everything else I have tried so far :-(

-- 
Mark Cullen <mark.r.cullen@gmail.com>