A storage system based on a disk array that holds a certain amount of redundant information. The redundant information can be used either to detect or in some cases correct errors. As with any system, the greater the fraction of the information that is redundant, the greater the protection against undetected errors, or the ability to recover from errors when detected.

In 1988, David A. Patterson, Garth Gibson, and Randy Katz of the University of California at Berkeley published a paper entitled ‘A Case for Redundant Arrays of Inexpensive Disks’, which outlined five array models or RAID levels. The levels were named RAID 1 through 5, although no hierarchical relationship was implied. Since the publication of the paper, a sixth RAID level has been described by the authors. In addition, RAID level 0 is used to refer to a stripe set (see stripe disk). However, the absence of redundancy in a stripe set makes the term RAID a misnomer. The use of the word inexpensive was because of the belief that arrays of low-cost PC drives offered a significant decrease in storage costs when compared to SLEDs (single large expensive disks), which at the time were used on mainframe systems.

Four RAID levels—0, 1, 3, 5—have been found to be commercially attractive; however, each has drawbacks when applied in products. RAID product developers frequently improve upon the data mapping and redundancy protection models outlined in the original paper. This is achieved by combining RAID levels and\\or combining RAID data mapping with other technologies such as caching. The RAID level 0 and the Berkeley RAID levels are as follows.

RAID 0

or Disk Striping Data is distributed uniformly in chunks across the member disks of the array; no redundant information is generated. If there are N disks in the array, its MTBF is 1/N times the MTBF of a single disk. The data transfer capacity and I/O rate is very high for both reads and writes.

RAID 1

or Mirroring All data is duplicated across the N disks of the array so that the virtual disk has a capacity equal to that of one physical disk. For N > 3 this configuration has the highest data reliability. The data transfer rate is higher than a single disk for reads and slightly less than a single disk for writes. The I/O rate is up to twice that of a single disk for reads and similar to a single disk for writes.

RAID 2

Each sector of data is divided into small chunks and is distributed across the k data disks. The virtual sector size is thus k times that of a physical disk. Data is protected by a Hamming code; the N disks of the array comprise k data disks and m redundant disks such that

$$N\le 2^m-1\text{and}k=N-m$$

Data reliability is comparable to RAID 3, 4, or 5 while the data transfer capacity and I/O rate is comparable to RAID 3.

RAID 3

or Parallel Transfer Disks with Parity Each sector of data is divided into small chunks and distributed across the N − 1 data disks. The virtual sector size is thus (N − 1) times that of a physical disk. The redundant information is stored on a dedicated parity disk. Data reliability is comparable to RAID 2, 4, or 5 while the data transfer capacity is the highest of all.

RAID 4

or Independent Access Array Data is striped to chunks on the data disks, which are very much larger than the data sectors. Redundant information is stored on a dedicated parity disk. Data reliability is much higher than for a single disk—comparable to RAID 2, 3, or 5. Data transfer capacity and I/O rate is similar to disk striping for reads but significantly lower than single disk for writes.

RAID 5

or Independent Access Array with Rotating Parity Data is distributed as in RAID 4 but to all the disks of the array. Redundant information is interspersed with user data. Data reliability is comparable to RAID 2, 3, or 4. Data transfer capacity and I/O rate is similar to disk striping for reads but lower than a single disk for writes.

RAID 6

is as RAID 5 but with an additional parity disk. The additional parity is independently computed such than any two disks in the array can simultaneously fail and the array will still provide user data. Data reliability is the highest except for RAID 2 with more than three member disks. Data transfer capacity and I/O rate is similar to disk striping for reads and lower than RAID 5 for writes.