Re: raid10 redundancy

[Phillip Susi <phill@thesusis.net>]

keld@keldix.com writes:

> My understanding is that raid10 - all layouts - are really raid0 and
> then raid 1 on top.

Naieve implementations work that way, and this is also why they require
a an even number of disks with 4 being the minimum.  Linux raid10 is not
naieve and can operate with any number of disks >= 2.

> If you loose two disks you loose the whole raid in 2 out of 3 cases in a 4 disk setup.
> If it was raid0 over a raid1 you would only lose the whole raid in one out of 3.
>
>                raid1
>            raid0  raid0      
>            d1 d2  d3 d4
>
>
> gone d1+d2  survives d3+d4
> gone d1+d3  dead
> gone d1+d4  dead

Only for the naieve implementation that considers the whole raid0 dead
as soon as one drive fails.  Linux raid10 does not have this problem
because it knows that if d1 and d3 fail, you still have a copy of all of
the required data on d2 and d4.

The bottom line is that the exact layout doesn't really matter.  You
have two copies of all data and so you can be sure that the array can
survie any one failure.  An additional failure has a 1/3 chance of
killing the whole array ( for a 4 drive array ).  Of course, you could
say, have an 8 disk array with 3 copies, then it can survive any 2
failures, and has a 5/6 chance of surviving another failure.  On the
other hand, you only get 1/3rd of the usable capacity of those 8 disks.