RAID Test in the works

Hi, We've all seen lots of discussion in this group about the pros and cons of RAID disk systems. We've seen lots of opinion and anecdotal evidence, but so far nobody's posted anything definitive. I'd like to develop some more usable data, and I'd appreciate ya'lls input on designing a test of RAID. HP has kindly loaned my firm a HP Fast/Wide SCSI array (8-gig) for our testing purposes. I'm using Informix 5.03.UD4 on HPUX 9.04 and have begun working up a test plan. Part of the testing is already done, but I can still add or modify it if necessary. My system will consist of several "normal" (not Fast/Wide) SCSI disks and the RAID array. The RAID can be configured as 5 independent drives, as a RAID 3 and as a RAID 5. So far, here's what I plan: Place Informix code in normal SCSI filesystem. Place rootdbs, physical logs, and logical logs on separate normal SCSI raw disks (same controller, different spindles) Create equal-sized Unix filesystems on RAID and normal SCSI devices Create equal-sized raw partitions on RAID and normal SCSI for Informix dbspaces. For each of ([normal SCSI],[non-raid independent drives],[RAID3], [RAID5]) DO Run standard disk benchmarking tests in Unix filesystems for both normal SCSI and RAID devices. Will use Bonnie and iozone tests. Run following tests on newly-created files in dbspaces Load: Load a 100,000 row table from a flat-file Flatfile lives in normal SCSI filesystem Read: Unload this file to /dev/null Update: Update 10,000 rows of the table (non-indexed) Copy: INSERT into new_table SELECT * from old_table (both tables in same dbspace) Delete: Delete all rows in the table Between each test, stop and restart the engine to make sure the buffers are empty. DONE I'm including the non-raid, 5 independent drives as a separate option because I expect that the Fast/Wide SCSI should be faster than my existing SCSI. Running the tests with the fast drives but without the RAID should isolate out the hardware differences. Am I missing anything? I believe that this will cover most of the areas needed to look at the effectiveness of this implementation of RAID. If anybody has any ideas for additions, changes, or modifications, please post them. I'll post my results if anybody's interested. Thanks, Joe -- --------------------------------------------------------------------------- Joe Lumbley(jlumbley@netcom.com) Dallas, Texas ----------------------------------------------------------------------------