Well, that makes RAID level selection easy. The best way to get 5 GB (actually, 6) out of 4x 2TB drives with protection against a single drive failure is RAID 5. 3 and 4 are candidates, but they are not often used because 5 works better.
If one drive fouls the sheets, your array will be "degraded." When you put in a new, blank drive, the RAID array will spend many hours "rebuilding" and then be a good RAID5 again. No, I don't know the instructions for accomplishing this with your particular setup.
Parity: "A literary or artistic work that imitates the characteristic style of an author or a work for comic effect or ridicule." Oh, wait, someone else has the Forum Jester title.
Parity is a relatively simple way to recover from the loss of one bit / byte / block out of a set of N. I will illustrate the trivial case of three data bits and one parity bit.
Data Parity Bit
000 0
001 1
010 1
011 0
100 1
101 0
110 0
111 1
Let's say you send me the message 100 110 111. You add parity and send me 1000 1100 1111. I lose two bits, and get the message 10x0 1100 1x11. It's easy for me to figure out that the missing bits were 0 and 1, looking at the overall parity (number of ones) in each overlong "word."
How does this apply to storage? Conceptually, in RAID5, you store one bit of each three-bit word on a separate drive, and the parity on a parity drive. Which drive is the parity drive alternated between three-bit words.
If any one drive dies, I can easily 1) Reconstruct what the message was for any given 3-bit word, and 2) Reconstruct the drive that died from the other three.
In reality, parity is not done at the bit level, but the block level, and block size can be configured.
(Yes, all of you out there, I know that a flipped parity bit will be better, so that there are always an odd number of ones. I'm being incredibly simplistic.)
Edit: Hah! I figured the parity wrong in my example message!