Some of the most common misconceptions about RAID are:
Most RAID-documentation claims that RAID 3, RAID 4 and RAID 5 each require at least 3 drives to implement. I say that all these can be implemented correctly with 2 disks. Think about it. Put aside your prejudice. What does RAID4 on two disks look like? One disk data, one disk parity. The generalisation to RAID3 or RAID5 is trivial.
The obvious response is 'that sucks! RAID1 is much better if you have just two disks!'. That is of course true, but that doesn't change that RAID 3, 4 and 5 are in fact well-defined for just two disks!
To end this, let's do a thought-experiment. Suppose I have one disk filled with data. I replicate this data onto:
I then hand you these arrays, labeled A, B and C in no particular order. Answers in e-mail please: how to tell which of these 3 arrays (A, B and C) are the RAID1, the RAID4 and the RAID5 array. I'm looking forward to your answers. [Since somebody asked: we are assuming that the RAID-implementation does not label anything. Since somebody else asked: I am giving you the raw disks, to look at with any tools you have available. I will also give you the original data disk, for any comparisons you might want to do.]
This subject groups a set of RAID 3/4/5 misconceptions:
The correct answer for each of these cases boils down to: the RAID3/4/5 specifications do not tell us how many disks to use exactly. You can use anything from 2 to an infinite amount, and the RAID3/4/5 space overhead will always be exactly one disk.
The downside of this defined overhead is that RAID3/4/5 only protect against failure of one disk. In a 20-disk array, the chance of a random disk failing within a certain time is much higher than in a 5-disk array, which is why RAID3/4/5 on a 20-disk array is basically a terribly bad idea. (Of course, combining RAID-levels on a 20-disk array, or just splitting it up into several RAID3/4/5-arrays, is perfectly fine.)