This article is from the storage FAQ part1, by Rodney D. Van Meter with numerous contributions by others.
The original taxonomy of RAID levels was published in the SIGMOD paper by Garth Gibson and Randy Katz in 1988 (see below). The taxonomy roughly classifies RAID architectures according to the layout of data and parity information on disks. It is NOT gospel and does NOT cover every possible architecture (it has been pointed out here that that would require an N-tuple showing data block addressing, number and types of parity and ECC information, etc.), but when used properly provides a vocabulary and establishes a framework for discussion. Raid Level 0 - Striping - Data is segmented and split onto multiple spindles. Short Reads - Easily handles multiple simultaneous reads Long Reads - Single operation can be split and processed in parallel Short Writes - Easily handles multiple simultaneous reads Long Writes - Single operation can be split and processed in parallel Redundancy - None Cost - Good (no extra hardware) Raid Level 1 - Mirroring - Duplicate data is kept on multiple splindles Short Reads - Faster (shorter latency) since resolution can be from any of multiple disks Long Reads - Faster since resolution can be from any of multiple disks () Short Writes - Slower since need to write to multiple disks Long Writes - Slower since need to write to multiple disks Redundancy - Excellent Cost - Expensive - at least double the spindle cost Raid Level 3 - Data protection disk - mathematical ECC type code calculated from multiple spindles and stored on another spindle. Short Reads - Normal speed (i.e. 1x per-spindle rate) Long Reads - Normal speed Short Write - Slower due to re-calculating of ECC code (including reading from other spindles and the ECC write) Long Write - slightly slower due to ECC writes, but less reading required than in short writes (*) Redundancy - Excellent Cost - only slighly more than no redundancy options Raid Level 4??? similar to 3, with block striping instead of byte. Raid Level 5 - Striping plus data protection - stripe data across multiple spindles (as in RAID Level 0) and have data protection calculations (as in RAID level 3) but don't put all the calculated figures onto one spindle, but spread it out. Short Reads - Normal Long Reads - Faster due to parallelism Short Write - Slower due to ECC calculation (including reading and writing) Long Write - slighly slower due to ECC writes (**) Redundancy - Excellent Cost - only slignly more than no reduncancy options ( should be the same speed as a single spindle) (* -- should be faster than a single spindle due to parallelism on write? somebody help me out --rdv) Benefits of RAID: High data availability (ie, if a single spindle crashes, no data is lost) Increased disk connectivity per system - since multiple spindles appear as one spindle to the computer system. Large capacity storage in a small footprint - Flexibility through intelligent array controllers Performance enhancements in some circumstances. Streamed or Streamified RAID??? (SHMO)