This paper was published by Patterson, Gibson, and Katz in 1988.
The motivation behind RAID was the slow improvement in disk performance in comparison to CPU improvements.
Because of Amdahl’s law, if 90% of the time of a program was seeking for data, if CPU improved 2x, but data seeking didn’t, then the program would only be 5% faster.
Instead of using expensive high end disks, the authors propose using cheap, inexpensive disks for storage. Writes may write data to more than one disk, which slows writes, in exchange for faster reads. However, reliability also increases, because there are many copies of the same data.
Raid 0 is a scheme where data is striped across the disks without any increased redundancy.
Raid 1 refers to mirroring the same data across two disks. Reads and writes are both slower under this scheme, but reliability is improved because the system can tolerate failure of one of the disks.
RAID 2 involves using error correcting codes (ECC) for redundant data. This isn’t as popular nowadays, since the firmware of most disks involves handling for corrupt data (fail-stop mode), where if a read is issued to corrupt data, it provides no data at all instead of corrupt data. Nonetheless, it is a way to provide in software what might not be guaranteed by hardware.
RAID 3 is similar to RAID 2, except it uses a parity XOR bit to store data. In the case of a disk-failure, the controller can reconstruct lost data from parity data.
This is extremely inefficient for small writes, since they have to read a sector from disk, modify it, and then write back the entire sector.
RAID 4 is the same as RAID 3, just that it uses striping in contiguous sectors, which makes serial reads and writes a bit better.
RAID 5 spreads out the parity sector across the disks in a round robin manner – so each disk has some data and some parity bits. This helps improve serial reads and writes, and improves redundancy, since losing one disk is fine most of the time.