[SATLUG] RAID 5 setup

[Brad Knowles]

On 1/19/08, toddwbucy wrote:

>  Ill post the benchmarks in the next day or so....hmmm by the way can
>  anyone recommend a good benchmark suit for a quad core. lmbench is slow
>  and unreliable on SMP boards.

For disk benchmarks, I usually start with the list at 
<http://www.acnc.com/04_02_01.html>.  I know Russell Coker has done 
some good things with Bonnie++ (I've been working with Russell for 
years on various benchmark programs he's responsible for), but you 
may also be interested in IOzone, IOBench, and/or IOstone.  Note that 
IOBench is an official part of the SPEC 2.6 benchmark toolkit.

For large file I/O, such as you'd be doing for video production, at 
least some of the above benchmarks may need some tweaking or special 
configuration to focus on the kinds of file sizes and distributions 
that you're going to need.


If you're going to give Bonnie++ a try, make sure to use the latest 
"experimental" 1.93d version (see 
<http://www.coker.com.au/bonnie++/experimental/>), and not the 
official 1.03b version.

I give Russell a lot of credit for all the hard work he's done over 
the years, but a lot of his best stuff has not yet been brought into 
the official "stable" version.  You may also want to drop him a note 
and see if he's got anything newer than the 1.93d version on his site 
-- he tends to do a lot of work in private and not update the site 
for long periods of time.


If I had to pick two benchmarks to try, I'd go with Bonnie++ and 
IOBench.  Bonnie++ is pretty well-known and not hard to get running, 
although it can take some work to figure out what the most 
appropriate testing parameters are for your particular situation. 
IOBench is the official disk benchmark tool component for SPEC, but 
I've found it's harder to get built and installed, although it's 
pretty easy to run once that's complete.


Beyond that, no artificial benchmark can begin to compare with your 
real-world load.  If possible, try using your real applications on 
real data files, doing the sorts of things you'd actually do in the 
real world.  Then re-configure the RAID array and try again.  Find 
out what really works for you, and compare that to what the 
artificial benchmarks might tell you.

-- 
Brad Knowles <brad at shub-internet.org>
LinkedIn Profile: <http://tinyurl.com/y8kpxu>