There was a discussion in the RAID development mailing list about booting off mirrored root filesystems in July 97. The problem is reintegration of a failed disk. To repair it, Linux must be up first...
There seem to be two "good" ways of doing this:
1. modify LILO to handle this
2. As suggested by Chad Robinson on July 23, 1997:
1) Power up 2) First active SCSI drive gets hit by the controller to boot. You've already installed this `manager' on each drive in your system. 3) System examines state information on the drive and determines its status. From this it should be able to tell quite a bit. 4) If there are errors, it is HERE that recovery tools should be employed, but obviously you don't have Unix up yet to do it, and we've already seen that many (if not most) people don't want a tiny Unix collection (/bin, /lib, and some other things) to be static - among others, I want those mirrored too! So you build them into the utility. If you examine LILO, it has a good deal of code in it that isn't really relevant to booting a high-availability Linux system. If you have an HA Linux box, when do you ever boot Windows or OS/2? At the most you want to be able to manage multiple images, but not much more than that.
The last point is important: on a HA Linux system, users are not very likely to also boot other operating systems.