RAID 10 - Probability Of Two Disks In Any Sub-Array (Mirrored RAID) Failing

[jhigham]

Chart illustrating the probability of a two drive failure in a RAID 10 (striped RAID 1 sets) configuration being a mirrored pair.

In other words, if any two drives fail simultaneously, the chart below illustrates the probability that the two drives are a mirrored pair (RAID 1) of each other, thus resulting in complete loss of data.

X axis = Total number of drives in RAID 10.

Disks Probability Of A Two Disk Failure Both Being In Any Sub-Array (Mirrored Pair)
4 33.3%
6 20.0%
8 14.3%
10 11.1%
12 9.1%
14 7.7%
16 6.7%

 

[Ericloewe]

I have to disagree with this interpretation. Yes, the numbers are valid, but they're conditional to a dual failure. That's meaningless, because a dual failure increases in likelihood as the number of vdevs grows.

 

[Arwen]

I'd have to agree with @Ericloewe, if you started with 2 disks in a Mirror vDev, then grow it by adding pairs at a time, the chances of both older drives failing at the same time is higher. Especially during the replacement of the first failed drive.

If I was building a ZFS Mirrored Pool, I'd try to move the disks around as I added more. Meaning instead of just adding 2 disks as a new Mirrored vDev, I'd replace one of the older ones with a newer one, (unless the sizes differed). Then make the new Mirrored vDev from 1 old and 1 new disk. Complicated, and still requires you to have backups.

 

[rs225]

You reduce the odds of dual failure by regular scrubs, SMART tests, and notification. Instead of failures catching you by surprise, you see them coming.

 

[wblock]

and still requires you to have backups.

Much like every other form of redundancy. :)