> David et al., > I don't know if it applies to Linux, but I vaguely remember a > RAID 1+0 where > the data is both striped and mirrored. Is this superior to > RAID 5?
Debateable. It's certainly hardware-hungry (whatever price/performance you win with the RAID 0 striping, you immediately hand back with the extra disk and chassis cost for mirroring vs a net loss of one drive slot for RAID 5) and since the disks are probably in the same chassis for speed of light reasons, you'd have to make a really convincing case that it is better than just paying attention to the alerts from your drive error monitors and replacing the first disk failure either preemptively or Real Soon Now after it goes. I always spec at least one hot spare drive in any RAID array I build so I have a few days to get a new one if one goes sour on me. The 1+0 arrangements I've seen in practical situations were all for speed purposes on arrays with very large disks. If you have a slow controller or a multichannel controller with a full load on all the channels, 1+0 is probably faster because the data management is much simpler. > Also, for > the very paranoid, I would guess that one could use a RAID > 5+0 where the > data is striped w/parity like RAID 5, then each RAID 5 volume > is mirrored. > This would seem very excessive, but very safe. See above. You've just moved the failure point to the chassis, unless you put the mirrors in another cabinet, which would substantially slow down the performance of the array. That's why the telcos use completely duplicated Y-cable arrays, often in separate rack strings. No common failure points. -- db
