[kwlug-disc] Should drives in a raid perform alike?

unsolicited unsolicited at swiz.ca
Fri Apr 30 12:19:35 EDT 2010


It would be interesting to know:
- both drives purchased at same time? (i.e. Should be relatively close 
in internals/firmware)
- is OS drive on same controller?
- if you reverse the connectors of the data drives on the controller 
do the stats follow. (I'm wondering if busy on OS with 1st drive, done 
by time of 2nd drive, but you'd think the stats would be independent 
of that. Which might leave the controller order as a factor.)
   - Not sure you could just swap them safely. Drive 'letter' changes, 
etc. fstab may be at UUID, which would be good in this instance for 
less likelihood of problem, but I'd also guess the RAID is set at a 
lower level / before that.

Richard Weait wrote, On 04/30/2010 11:48 AM:
> Given two disks in a RAID0:
> - They are nominally identical disks, same manufacturer, same
> capacity, same part number
> - each has only a single, full disk partition (sdb1, sdc1)
> - those partitions are RAID0'd together (md0)
> - both drives connect to the same drive controller
> - the operating system is held on a different drive.  (sda)
> 
> Would you expect drive performance to be noticeably different between
> the two?  I didn't expect that.  Munin (stats graphing tool) reports
> 
> drive | iowait(ms)| util
> sdb  |  253.66   | 43.19%
> sdc  | 151.20    | 38.40%
> 
> Which suggests (?) that sdc is done faster than sdb.  Shouldn't they
> be doing the same thing in the same time?  The graph makes this look
> worse during the load on the left side of the graph.
> 
> Visual aids - munin graphs of disk performance during heavy load.
> http://imgur.com/a/LP20V/raid_vs_drives
> 
> Thoughts?  Am I expecting too much similarity between drives?



More information about the kwlug-disc_kwlug.org mailing list