post #11 of 11
RAID1E still won't do the double redundancy you are looking for, and neither will RAID5. Which is the point I have been trying to make all along...

RAID1E writes 2 copies of the data across 3 drives, making sure each copy is on a different drive. So you get half the capacity of your 3 drives. Consider the data ABCDEFGHI. This gets written:

Disc1: ABDEGH
Disc2: ACDFGI
Disc3: BCEFHI

The setup will survive 1 disc failure, and a second failure will lose all data.

For RAID5, you don't write 2 copies of anything. Instead you calculate a parity bit from your data, and write that to the drives in turn instead. So your data is written like this:

Disc1: _A_._C_.Pef._G_._I_
Disc2: _B_.Pcd._E_._H_.Pix
Disc3: Pab._D_._F_.Pgh._x_

Where Pab is the parity calculated from A and B, Pcd is that from C and D (other characters are just for spacing).

The problem with both of these RAID levels is they are very slow - RAID5 has terrible small writes performance, and RAID1E has (relatively) poor performance all round.

It is likely your best bet is to run your drives separately. Or if you decide you need redundancy then get a fourth drive and run 2 separate RAID1 pairs with your VMs split between them (that is how I run my server - I get much better performance like this rather than with a single RAID10 array).