RAID 0: Also commonly known as a stripe set (as it provides no redundancy, one of the useful parts of RAID) data is broken into blocks by the controller or the software and alternating block sent to each disk in the array. By doing this, sustained transfer rate is increased. However, as this has no effect on the seek time it does not improve desktop usage. It is useful mostly for video editing requiring tremendous STR. Also, MTBF of the array is decreased dramatically with each disk that is added, so array availability is very poor. This requires two or more disks.
RAID 1: Also commonly known as a mirror set, each disk contains an exact copy of the other. Because each disk contains exactly the same information, under a heavy random read load a mirror set can provide nearly double the performance as each disk can independently service a different request. However, they must execute a write together, which leads to modestly lower write performance. Two disks are required.
RAID 2: Also commonly known as hamming, uses a bit level data security algorithm. It is very compute intensive and not commonly used outside of IBM mainframes.
RAID 3: Also commonly known as striping with parity, this takes a RAID 0 array and adds an additional disk that stores byte level parity information on it, allowing data to be reconstructed following a disk failure. Read performance is similar to a RAID 0 array, however write performance is usually low due to the need to read from all of the disks before writing the parity information, and in sustained writes the need to write the parity information for each block to a single disk. Hence write performance is worse than that of a single drive. Three or more disks are required.
RAID 4: Also commonly known as striping with parity, this is identical to RAID 3 except provides block level parity information. Three or more disks are required.
RAID 5: Also commonly known as striping with distributed parity, this takes the parity information in RAID 3/4 and distributed it across all of the disks in a striped manner, allowing for higher write performance by removing the parity disk bottleneck, and nearly the same read performance. This is commonly used for massive file storage and some transactional databases. As with RAID 3 and 4, it can sustain a single disk failure. RAID 5 requires three or more disks.
RAID 10: Also commonly known as striped mirror sets, this is what you get when you take two or more RAID 1 arrays and stripe them together. Performance is usually outstanding, and it can sustain one disk failures or, depending on the controller, up to half of the disks in the array (providing them being the correct disks). Four or more disks are required.
RAID 01: Also commonly known as mirrored stripe sets, this is what you get when you take two RAID 0 arrays and mirror them. Performance is usually outstanding, and it can sustain one disk failures or, depending on the controller, up to half of the disks in the array (providing them being the correct disks). Four or more disks are required.
RAID 50: Also commonly known as striped RAID 5 arrays, you get this by taking two RAID 5 arrays and striping them for higher capacity and speed. Six or more disks required.
RAID 51: Also commonly known as mirrored RAID 5, this is what happens wehn you take two RAID 5 arrays and mirror them for higher security and performance. Six or more disks required.
Also of note: many hardware RAID controllers represent the RAID device to the OS and a SCSI disk, leading to a maximum of 2 TB per array (its a SCSI thing). When using combined RAID levels (10, 01, 50, 51), this can get around that limitation by using software RAID built into most modern operating systems.