[Laurence]

Does anyone have an explanation or even a good link as to how RAID5 works? I know that the array has the total space of N-1 drives assuming all drives are equal size. How does this really work? It seems to me that if I have 11 1TB drives making a 10TB array that there can't really be enough space to back up all the data on one drive. If that one drive fails, how is the data regenerated from the parity information? as I understand parity (from the days of 8N1 modem connections) parity can only tell you that every bit is correct -OR- some bits are wrong. What happens when multiple random bits are wrong. Can the parity tell you enough information to completely recreate the data? If this is the case it seems that no space at all would be required for parity on an N-1 array as N approaches infinity. I realize that this is unrealistic in that we don't have infinite drives to test, but I can imagine that even with a few hundred drives something has to break down.

[OreoCookie]

Have a look here. They also explain all other RAID levels.

[Laurence]

Thanks for the quick reply, but I actually understand the RAID levels reasonably well... What I don't understand is how, if one drive is missing in a RAID 5 array you can take the "parity" information and recreate the original file. If the parity information is, by design, smaller than the original data it would seem to violate some mathematical law. If the parity is enough information to recreate the original data then you could just store the parity data and not store the original files at all? If you repeat this process indefinitely you would have the "perfect" compression algorithm where you could compress anything indefinitely which is obviously impossible. I was looking for some information on the math behind the RAID 5 restoration process.

[mduell]

You lose 1 drive capacity to parity, so there's no mathematical violation. With four 1TB drives in a RAID5 array, you can store up to 3TB. If you lose 1 drive, you now have 2TB of data and 1TB of parity, from which you can reconstruct your 3TB of data.

If I recall correctly, RAID5 is XOR based. Say you have two data bytes, A and B, stored on separate disks. Now calculate A XOR B and store it as C on a third disk. What happens when you lose a disk?
Lose the disk with A on it: calculate B XOR C and you have A back.
Lose the disk with B on it: calculate A XOR C and you have B back.
Lose the disk with C on it: calculate A XOR B and you have C back.

At 11PM I can't remember or figure out how they do it with more than 3 disks, but it's the same idea.