OT: Mirrored or RAID5

[Glenn Steen]

On 02/12/2007, Koopmann, Jan-Peter <jan-peter at koopmann.eu> wrote:
> > R1 gives better performance than R5, for both
> > writes and reads. Combining striping (sometimes called R0) with
> > mirroring (called R1+0) will give you the best performance.
> > As a stopgap, you can have a large cache on your Raid controller, in
> > which case small I/O loads will never really depend on the disks, so
> > then an R5 can be acceptable.
>
> I do not think this is true in all cases. It really depends on your
> hardware and what kind of data is read. If you have a really good RAID
> controller it will do the RAID5 checksums wire-speed. In that case
> having many hard disks can and will give you better performance than
> "just" two disks and a RAID-1 set, depending on the discs etc. Reading
> from 10 disks will give you quite a throughput. :-) Moreover RAID-5 is
> quite good in reading chunks of data in multi-user environments.
>
> I agree though: A 10 disk RAID-10 will probably outperform a 10 disk
> RAID-5 in most cases.
>
I have only confirmed this with EMC storage/controllers, so true, you
might be right that _some_ configurations of  R5 <> R1 will differ a
bit on reads. On writes though, R5 will likely lose.
And R1+0 will win the day any time. All depending on the amount of
spindles, of course.
For MS use, this is highly hypothetical, since MS hardly puts much
load on the disks.... Not like a very active DB;-).
As a rule of thumb, the "avoid R5, if you can" isn't far of the mark though.
But in Phils case, he has a situation where utilisation on one R1 is
really low, and then he might want to make it all one big happy R1+0
(speedy, but costly) or ane slightly slower R5.... Which might bd just
fine... I'm just sharing my views here:-).

Cheers
-- 
-- Glenn
email: glenn < dot > steen < at > gmail < dot > com
work: glenn < dot > steen < at > ap1 < dot > se