On Wed, 16 Jan 2008, Josh Paetzel wrote:

> RAID is no substitute for backups.  In 12 years of using RAID I've lost 
> arrays that required multiple drive failures twice.  Once was a trio of 
> PATA drives in a mirror with a hot spare, drive 0 failed, and while the 
> hot spare was syncing drive 1 failed.  The other incident was a RAID 5 
> array of SCSI drives with a hot spare.....a bad PSU left scorch marks on 
> all 4 drives.
>
> In any case, I realized the importance of backups a long time ago, and 
> while I can't say I've never lost data I can say it's been a couple 
> decades since I lost data I cared about.


These are examples of non-independence of failures.  In my earlier message 
I said that RAID1 fails catastrophically with probability p² under 
independence.  Under complete dependence it fails with probability p 
(which is a lot worse - just as bad as using only one drive).

I have to hope that RAID1 doesn't show complete dependence in practice 
because it would be useless if it did.  The real probability of 
catastrophic failure must be somewhere in between p and p².

Mike