Well, we can just start at the top and go through it.
The number 1 thing to do with either IDE or SCSI is to determine what RAID level you want to use. This even will help determine what type of card you want to buy.
Levels, with descriptions: (This is all out of my head at the moment, if I screwed one up, let me know)
0 - Striping
Striping takes the data you want to read or write and divides it into even chunks. For instance, if you want to write to a 128k block, that can either write 64k to two disks or 32k to four disks. You can also have an odd number of drives, but I won't go into that for the sake of math. All tasks are done to multiple disks at the same time, thus increasing performance by a factor of the number of disks, minus overhead. A good SCSI RAID 0 can pull a 90% improvment from two drives, and about 70% or so for a third. The return decreases as you go. The drawback is that if you lose any one drive, you lose all of your data. Also, access time tends to suffer a little. This requires a minimum of two disks, and any number of disks.
1 - Mirroring
Mirroring does just what it sounds like, it mirrors your data on multiple disks. You can mirror as many times as you'd like (up to the amount of drives you can support) depending on how paranoid you are. Because of the nature of what's being written and read, read operations can be accelerated by 100% (minus overhead) but write operations usually stay the same. This is the number one way to go for data protection. The only way to lose your data is if loss occurs from a fire or other external damage, or possibly, if you lose the controller and can't replace it with another one. Most of the time though, the single drive will still work just fine in any other system. This requires a minimum of two disks, and an even number of disks.
3 - Striping w/Parity
Striping with parity increases performance while maintaining a handle on redundancy. RAID 3 does this by implementing a RAID 0 and then creating a seperate disk to write parity information. This way, if you lose a disk, that disk's info can be recreated. This works on a binary system. (i.e. 11=0parity, 00=0parity, 01=1parity, 10=1parity) This way you can take any two bits, and recreate the lost one. The benefits of this are performance and safety, although with RAID 3 you put a large strain on the odd disk that contains parity as everything has to be calculated and written to it. For ever bit written to any other disk, one gets written there, both bottlenecking performance, and creating more strain on this disk. This is why 5 is considered to be a much better option. This requires a minumum of 3 disks, and an odd number of disks.
5 - Striping w/Rotating Parity
RAID 5 does the same thing as RAID 3, with one major difference. Each write operation, the parity is moved to another disk. This way the strain of writing parity bits gets passed around to all of them, instead of just one drive. Except for that you still get the same benefit in performance, an increase like RAID 0 (though not quite as fast due to calculating parity) plus the security that if you lose any one drive it can be replaced and the array can be rebuilt. It's not as good as RAID 0 though, because you can't easily just pull a RAID 5 array out and get it going on another machine without a bit of effort. This is the level most often used in business environments, because at a large scale you are only losing 33% (3 drive) 20% (5 drive) or even less of a percentage of the disks you purchase vs. 50% with mirroring. RAID 5 requires an odd number of disks.
10 - This is the hybrid between RAID 0 and RAID 1. RAID 10 is a mirrored stripe, and often is supported by less expensive cards, built on cards, Promise type cards, etc. This will lose you half of your storage due to the mirror. What you do get is increased speed due to striping and reading and writing to multiple disks at once, as well as the very safe nature of a RAID 1 system. This requires a minumum of 4 disks, and an even number of disks.
15 - Just like ten this is the cross between RAID 5 and RAID 1. You simply take two complete RAID 5 arrays, and mirror them. One huge benefit this gives, besides the obvious safety and the speed of RAID 5, is that if you lose a disk, it can be rebuilt much faster due to you already having an exact duplicate of it. This can be of a large benefit in heavily-utilized production environments where the hit from calculating each read due to a bad disk can cost a lot of time and money. This way downtime is minimized. This requires a minimum of 6 disks, and an even number of disks.
50 - RAID 50, the combo of RAID 5 and RAID 0. One major benefit here... SPEED!! RAID 50s are very cool, they take the data you want to write, say 256k, then split that among the RAID 0, so 128k + 128k, then split that among the RAID 5s, so you could be writing 32k + 32k + 32k + 32k + 32k + 32k + 32k + 32k all to separate disks at the same time. The same is true in reverse as well for reading. You could also lose 1 disk out of each array and the thing would keep chugging along. I have personally set one of these up, and it was quite nice. You could stream massive amounts of info to several machines at once over the network without it even breaking a sweat. The drawback of course is cost. To do it right it really should be done on two controllers, or one multi-channel controller to give the arrays as much bandwidth as possible. This requires a minimum of 6 disks, and an even number of disks.
These are the most common levels you will ever see in production, there are some odd weird ones, but anything that supports them is probably more for legacy replacements than new installations. If you have any questions feel free to ask and I'll try to answer them the best I can.
Hopefully this is a good chunk for the FAQ. 🙂