I have been fighting the urge of replying to this because I know I will get some yellins but here goes.
The described scenario is reproducible on PERC cards when operating in a noisy environment. Bad signal integrity will always get you and eventually knock some drives offline. The described mechanism of recreating the RAID set by entering the exact values as before and not initializing it is the best trick to recover from multiple failures. If only one drive fails let the rebuild take care of it because you get the added bonus of reading ALL disks and therefore remapping any sectors that might be going bad. One of the worst offenders I have seen with my own eyes was a supermicro that has a U320 backplane inside the chassis that was connected from a riser with a cable that was too short (yes there is such a thing in the SCSI spec). So lets recap; chip on motherboard gets routed over quite a distance to a riser card, from the riser card it gets routed on the riser board (I didn't get to look at the routing on that one but I wonder how well that was done considering the rest of the routing work that I saw) and then through a plastic connector to a cable that was roughly 5cm (at least 5cm too short) to a backplane where the drives plugged into. Again eyeing the other pieces in the chain I am not so sure about the quality of that one too. I was honestly horrified after seeing that. Just about every rule was broken on obtaining clean signaling. U320 took a few months to define and then 3 years to get the signaling right. Most of the development time in the U320 stack was spent on getting signaling to work right. This is one of the reasons U640 was abandoned; it simply would have been too expensive to go trough the same hoopla to get marginal speed gains on a parallel bus and hence SAS became the next SCSI version. The good news here is that the industry did seem to have learned a lesson. No more plastic cables and connectors that wiggle, no more cables that can't be bent at a normal radius, fully Faraday caged cables from connector to connector etc. SAS has a very good signaling package and is superior to parallel SCSI. And then they decided to plug SATA drives into SAS cards... Which means plastic and wiggles for everyone! History is repeating itself and manufacturers have been scrambling to create interposers that remove the SATA signaling unknowns. Interposers are those little boards that plug directly into a SATA drive and then are plugged into a backplane of sorts. This is done to extend the distance a SATA drive can talk (driver strength isn't spec'd right resulting in a lose interpretation by vendors), filter the noise out of the connection, use a quality shielded SAS cable instead of a crappy SATA one, provide 2 paths to the same disk (SAS disks have 2 paths by default) and a few more minor reasons. The idea is to remove most of the SATA badness (read cheapness) and stabilize the whole environment. Over time this has become better and better however there is only so much one can graft onto a deliberately cheap device.
