On Wed, Feb 04, 2004 at 09:38:49AM -0600, Bill Gradwohl wrote: > I find it difficult to trust "software" RAID from past experience on > other O/S platforms, Don't blame Linux for the sins of other OSs. > and [I don't trust software raid] from the purely logical > perspective. Why not? When the system wants to read it asks the disk that appears positioned closest to the desired data, and when the system wants to write it issues overlapping writes to each of the controllers with raid disks. Fast and conceptually simple. I once had a disk die in the middle of burning a CD, and I never noticed a thing was wrong until the next morning when I got an e-mail telling me I lost a disk. (This was RH 7.2, I had set up a cron task to watch for changes in /proc/mdstat, and then send me e-mail if it found any.) Linux software raid 1 is even bootable--though getting your BIOS to try more booting from than one disk might be tricky: but always be careful about attempting remote reboots of degraded machines. > not as good as "real" RAID. One key point of raid 1 is reliability, and reliability likes simplicity and eliminating single points of failure--and getting things back up and running quick when they do fail. If you do software raid 1 and something goes wrong, all your parts can be inexpensive, generic, and readily available. In fact they can be cheap enough to keep spares on hand. If, instead, you buy a "real" hardware raid controller you now have an expensive, specialized, and rare item--the raid controller--which has complexity and can itself fail. Then what happens? Do you have another in stock as a spare? Do you expect your local Microcenter to have a replacement in stock? Certainly hardware raid has its place, but probably only if you have redundant machines and you can seamlessly ride out the week or two it might take to replace any specific spare. And by then you are probably doing SCSI or fiber channel disks anyway; I don't pretend that what I am saying applies in that region of larger installations. > Software RAID is corruptible by definition. By that token all "software is corruptable by definition", isn't it? Are you going to quit running any software? I am far more afraid of disks that might not flush written data all the way out to the platter than I am of the Linux raid code misbehaving. Finally, when comparing "hardware" to "software" remember that these days nearly all "hardware" of significant complexity actually includes a CPU, RAM, and software in some sort of ROM. The software in Linux raid code is very well tested by now--will that be true of the embedded software in your "hardware" controller? You also presumably have a procedure for keeping your Linux code up to date--will you have as good a way to keep track of any updates for you raid firmware? Will needed firmware updates be as promptly released? -kb, the Kent whose /dev/hda and /dev/hdc raid disks are not world readable nor writable.
