[lug] RAID under Linux

D. Stimits stimits at idcomm.com
Sat Mar 24 01:58:49 MST 2001

"J. Wayde Allen" wrote:
> Has anyone tried using software RAID under Linux?  If so, how well does
> this work?
> I'm setting up a new server for my lab and was thinking about trying
> it.  Basically, I'm wondering if there are any words of wisdom or gotchas
> that I should be aware of?
> - Wayde
>   (wallen at lug.boulder.co.us)
> _______________________________________________
> Web Page:  http://lug.boulder.co.us
> Mailing List: http://lists.lug.boulder.co.us/mailman/listinfo/lug

I found it works quite well with the 2.2.x kernels, haven't used it
since I recently updated that machine and did a total wipe and reinstall
(it was basically a test machine; most of the wipe was removing the
windows partitions it used to have). The version I ran was the RAID 0
(striping, versus linear append), all scsi. You probably want it
directly compiled into the kernel if you think you are doing more than
testing out the water. Another consideration is that if two drives or
their partitions are not fairly closely matched, you can lose a
significant amount of performance. I think that if you use two different
brands of drive with exactly the same geometry and specs, you might
still lose a small amount of performance. When you start dealing with
something like a small drive combined with an equal size partition from
a much larger drive, you'll lose fairly significant amount of
performance. And if you use IDE on the same cable, well, don't even
bother. There is also higher cpu load with software RAID; dual cpu or
more cpu's can actually outperform some of the hardware RAID though.
Basically if you have a slow cpu now, I wouldn't use RAID if it is for
performance reasons (unless of course you have dual slow cpu, then it
can work wonders). Do also read up on chunk sizes so you get a good size
for you, it makes a large difference on performance if you can choose
that wisely for small files or large files, so on.

D. Stimits, stimits at idcomm.com

More information about the LUG mailing list