Raid Array Stripe Size Investigation

[Rishi Chopra]

After some initial trouble getting my FreeBSD box up and running, I'm 
happy to report that I've been able to conduct a few empirical tests on 
IDE hardware RAID array stripe size and performance.  I decided to do 
these benchmarks when I noticed a lack of this information on the web 
and as part of the mailing list.

Information on the setup of the tests and results can be found here:
http://www.ocf.berkeley.edu/~rchopra/RaidResults.html

If anyone would care to add some insight into why Bonnie and IOZone 
are/aren't good benchmarks for such a test, feel free to share.  Also, I 
didn't bother to write any conclusions to the tests; I merely generated 
the data.  As with everything RAID, there's no 'best' configuration, and 
it doesn't seem like there was a clear winner as a result.  I also doubt 
whether you could draw any conclusions about the OS and hardware driver 
implementation. As far as single-user systems go, I don't think you'd 
see any difference performance-wise regardless of stripe size.  If 
someone can suggest a methodology for multi-user testing, I'd consider 
giving it a go.

NOTE:  I do still have a question about large disks.  My 4x200 RAID5 
array (~550GB) won't load if I tell sysinstall to use the entire disk as 
one slice, with partitions amongst that slice.  Given the large size of 
the array, I ignored warnings about '# of cylinders' but I'm still 
curious why I can't use the whole disk as one bootable slice.  Any 
suggestions?