Hosting again

[Simon Burr]

On Tue, Oct 30, 2007 at 12:17:05PM +0000, Aidan Samuel wrote:
>On 10/29/07, Simon Burr <simes at bpfh.net> wrote:
>> RAID 0+1 and 1+0
>So, er... is there a difference?

Very much so; the ordering is important. With 0+1 you mirror stripes and with
1+0 you stripe mirrors. A little bit of ASCII art can help here; we have six
disks, A..F

          0+1                                 1+0
         A   D                               A===D
         |   |                                 |  
         B===E                               B===E
         |   |                                 |
         C   F                               C===F

A + B + C form a RAID0 set ABC       A+D are mirrored; AD
D + E + F form a RAID0 set DEF       B+E are mirrored; BE
ABC and DEF are mirrored.            C+F are mirrored; CF
                                     AD + BE + CF form a RAID0 set.

One benefit of a 1+0 over 0+1 is how it copes with failed disks. If in 0+1
disk A failed then the stripe of A+B+C would need to be invalidated. If
you lost another active disk - eg D - then that would be it, you'd loose
the other side of the mirror. For 1+0 if you lost disk A then only one half
of the AD mirror is invalidated; the RAID0 set as a whole is still okay. So
long as you don't loose both halves of any given mirror then you can loose
as many disks as you have mirrors - eg you could loose A, B and C without
data loss. What is often done is to put one half of the mirror into one
disk array which is connected to one SCSI controller and the other half in
a different array on a different SCSI controller. You can then survive the
loss of the SCSI controller, route from controller to disk pack, disk pack
hardware, etc.

The rebuild times are different as well - in 0+1 you'll have to rebuild the
entire A + B + C stripe from the other half of the mirror. For 1+0 you only
have to re-build the AD mirror. For larger gaggles of disk - eg for 12 drives
on each half - this starts to become quite important.

-- 
   Simon the stressed        http://www.bpfh.net/           simes at bpfh.net
                 Chocolate is *not* a substitute for sleep