[Gllug] Production system - Linux 2.4.24, LVM and cciss

[Mike Brodbelt]

On Sun, 2004-01-11 at 21:14, Dale Gallagher wrote:
> > Almost by definition RAID5 is faster for both reading and writing than
> > RAID0+1.  In cases where RAID5 is actually slower, it often means that
> > the system was poorly thought out.  For example, small numbers of
> > larger disks where larger numbers of smaller disks would been
> > appropriate.
> 
> I disagree - in fact, I haven't found a single comparison where RAID5's
> overall performance is claimed to be better than RAID1+0 (I've been
> talking about a stripe of mirrors [1+0], though a mirror of stripes
> [0+1] shares the same level of disk performance). I've included a few
> references:

It's one thing to talk about RAID level performance in tests, but in the
real world, it's not hard to build an array to perform to a specified
level using either RAID5 or RAID10. RAID performace is primarily based
on the number of spindles in the array, with bus capacity also making a
difference, so in some cases you will want your spindles spread over
multiple IO channels.

RAID5 buys you more space, RAID10 a little more redundancy. There was a
time when the parity calculations needed for RAID5 slowed its write
performance heavily, but modern CPU speed and a caching controller makes
this immaterial today, IMO.

Mike.

-- 
Gllug mailing list  -  Gllug at gllug.org.uk
http://lists.gllug.org.uk/mailman/listinfo/gllug