Assuming MD exists between the device layer and the file system layer,
how do you ensure that important information (like copies of
superblock) is stored across the disks? The way you (S.Shapiro)
describe it, if you store a copy of the superblock every 2^n blocks,
there is a fair chance all the superblocks get put on *one drive only*
with striping, and when this drive gets...uh...run over by a truck or
something, salvaging the file system is going to be difficult. Is
some measure taken to prevent this?
There's also the performance issue, but that appears to be slightly
more complex, so I won't even try :-)
-kzm