true… kinda got lost in the speculation … xD
to get back to the point… i would say clustering of storagenodes, might require some sort of software.
but really 95% is down to regular VM clustering and running onto separate storage solutions running some kind of raid.
IMO, only issue so far as i see it, is like you mentioned a kernel panic in a VM would essentially shut it down, and then it would be practical if a storagenode could figure out utilizing multiple identities on multiple locations depending on if it detects that other local storagenodes crashed…
tho CentOS without to many bells and whistles is rock solid… and i suppose in case of a kernel panic one could set the IPMI to simply reboot the system using the watchdog features… i believe…
so yeah… i think all the stuff for it exists without software changes… or minimal in an case to take care of highly special cases of crashes maybe…