[ale] "Cost" of raid?

Danny Cox danscox at mindspring.com
Fri Sep 27 08:01:42 EDT 2002


Robert,

On Fri, 2002-09-27 at 06:43, Robert L. Harris wrote:
> Nope, can't aquire any new hardware til next budget.

	Where have I heard THAT before? ;-)

> Could anyone say that writes to Raid 5 take 1.25x as long, 4x as long,
> etc?  When the array looses one disk, reads take 5x as long as 0+1,
> writes take 7x, etc?

	Someone may say, but they'd be wrong!  This, as in most things, depends
on many factors.  

	For example, the SGI XFS fs (which we use here) was quite a bit slower
on RAID5, due to the way it writes it's log file in smaller chunks. 
When a log write comes along, the software RAID5 driver flushes all of
it's stripe headers for the "old" size, and reallocates them for the
"new" size.  Then, it switches back.  Recently, though, SGI has made
some changes in the way they write the log which helps greatly.

	Regardless, RAID5 will typically write slower than RAID1 (even though
you're writing the data twice in R1).  The RAID5 driver uses every trick
I've ever heard of (and one new one that I'd not heard of ;-) to speed
itself up.  It's very good.

	Another point of view is: can/will you sacrifice some speed for the
capacity?  The size of a RAID5 device is (size_of_drive * (# disks -
1)).

	That's just a for-instance.  Since your milage will vary, you have to
try several different combos using your particular mix of activity.

-- 
kernel, n.: A part of an operating system that preserves the
medieval traditions of sorcery and black art.

Danny


---
This message has been sent through the ALE general discussion list.
See http://www.ale.org/mailing-lists.shtml for more info. Problems should be 
sent to listmaster at ale dot org.






More information about the Ale mailing list