Benching RAID and AX100 vs. single SCSI disk - weird results

Fran Fabrizio fran at cis.uab.edu
Tue Mar 22 09:37:17 CST 2005


I'm currently evaluating an AX100 for purchase.  I hooked up the AX100 
to a RHEL3 PE1750 with a QLogic qla200 HBA directly (so, In a DAS 
configuration).  I created a single LUN, a RAID5 partition across the 4 
250GB SATA disks that are in the AX100 currently, and formatted it as 
ext3 (most of the things I will compare it to also have ext3).  I ran 
some Bonnie++ benchmarks on it vs. a single SCSI disk and some other 
RAID configurations I have on my network.  I saw some very surprising 
(to me, anyway) results:

- The single SCSI disk outperforms all of the RAID solutions in most 
categories
- I saw some really weird results where a 3-disk RAID5 on a PERC blew 
away every other solution on file creates/deletes per second
- Under load (4 concurrent Bonnie++ instances), the AX100 degraded as 
much as a single SCSI disk (contrary to what our Dell tech rep was 
expecting to see)

In general, I'm surprised that RAID in all of the various configurations 
I have cannot outperform a single disk (particularly for reading).  
Also, I'm surprised that the AX100 is playing in the same ballpark as 
other RAIDs and even NFS for some tests, whereas I was expecting better 
performance from it.

Some people who have already seen my results have said that it might be 
due to Dell RAID controllers being slow, and I've definitely heard that 
chatter on this list in the past.  However, I think there's something 
more going on (such as the AX100 is misconfigured, or I have 
unreasonable expectations of RAID and the AX100). 

Anyhow, my results are at http://www.cis.uab.edu/fran/ax100-3-21.pdf

Page 1 - a single Bonnie++ run
Page 2 - single run but with OS buffering disabled (forces an fsync 
after each I/O)
Page 3 - 4 concurrent runs (only AX100 vs. local single SCSI disk, 
because we were expecting to see the AX100 handle it better than the 
local disk, but nope)
Page 4 - some notes on the different storage systems I tested and how I 
ran Bonnie++

Any and all thoughts as to how to explain what I am seeing would be much 
appreciated.

Thanks,
Fran

-- 
Fran Fabrizio
Senior Systems Analyst
Department of Computer and Information Sciences
University of Alabama at Birmingham 
http://www.cis.uab.edu/
205.934.0653




More information about the Linux-PowerEdge mailing list