[NTLUG:Discuss] Built a big box

[Kelledin]

On Friday 24 January 2003 12:04 pm, Dan Carlson wrote:
> I'm always interested in disk and filesystem performance.
>
> What do you get with hdparm -t on your raid partitions?
<snip>
> Yes, I realize hdparm -t is a simplistic low-level synthetic
> benchmark, so it isn't a good way to measure realworld
> performance.

Problem is, simplistic benchmarks like hdparm -t show some RAID 
arrays in their worst light.  storagereview.com discovered that 
about WinBench a few years back when they started benching RAID 
setups--the sustained xfer rate (one of the first features a lot 
of people look at first in WinBench disk tests) was actually 
_lower_ than that of a single drive.

I confirmed that for myself when I ran WinBench on my own RAID 
array (10xQuantum Atlas 10K II, RAID50).  Sustained xfer rate 
appeared to be subpar, while other real-world tests like 
WinBench's photoshop subtest were through the roof.

> If you want to run bonnie tell me which command line
> parameters you want to use.

Definitely bonnie would be the thing to use for Linux.  IOmeter 
would be the disk benchmark of choice for Windows; 
storagereview.com switched to that after the mess with WinBench 
and RAID.

I've heard lmbench makes a good general-purpose benchmark, but it 
segfaulted last time I tried it.  Don't know why.

dbench makes an excellent stress-test, and there's the 
ever-popular kernel compile run. ;)

Also, MySQL comes with a benchmark/testing suite that could be 
used here.  The output's a little fugly, and the tests sometimes 
give false alarms, but it's usable if you massage the script a 
bit.

-- 
Kelledin
"If a server crashes in a server farm and no one pings it, does 
it still cost four figures to fix?"