vdev/pool math with combined raidzX vdevs...

[Adam Nowacki]

On 2012-07-06 03:16, Jason Usher wrote:
> It's easy to find the failure math for raidz2 and raidz3.
>
> But what if you create a pool with 3 different raidz3 vdevs inside of it ?
>
> For instance, 3 12-drive raidz3 vdevs in one big pool.
>
> For each individual vdev the failure probability is now higher, since not only will it fail when 4 drives in the vdev fail, but it will also fail if four drives in any of the other two vdevs fail.
>
> So each raidz3 vdev now has a failure rate higher than vanilla raidz3 ... but what is that new failure rate ?  Is it still higher than vanilla raidz2 ?

If you're asking if 3-stripe x 12-drive raidz3 has a higher failure 
probability than 36-drive raidz2 then the answer is no. It also doesn't 
have a higher failure probability than 36-drive raidz3. But if you 
compare 3-stripe x 12-drive raidz3 with no stripe 12-drive raidz2 then 
for a disk failure probability less than ~0.16 raidz3 wins, otherwise 
raidz2.