RAID in its several forms provides the backbone for most of our high availability, high performance storage. RAID devices have been with us since the late-1980s, and by now are so much a part of our IT lives that many folks have even forgotten what the acronym stands for.
Actually there have been several meanings for RAID over the years. The most frequently used has been Redundant Array of Independent Disks, but the "I" has also stood (often, inaccurately) for "inexpensive," and the "D" has on occasion meant "devices" (useful when describing that relative rarity, tape arrays).
Many levels of RAID have been defined over the years, and most have been implemented at one time or another. Most prominent among these have been:
* RAID 0, which stripes data across its disks to provide faster access but provides no added protection.
* RAID 1, which duplicates data via mirroring to provide fault tolerance.
* RAID 3, which offers speedier reads by striping data across the drives and then accessing the drives in parallel. This offers fault tolerance because it stores parity bits on a separate, dedicated disk drive.
* RAID 5, which stripes both data and parity across all drives to get better load balancing. This in turn provides better transfer speeds on reads along with increased data protection.
All of these RAID levels were defined by an industry organization, the RAID Advisory Board.
There have been other RAID implementations as well. RAID Levels 2 and 4 were defined by the RAID community, but were rarely used and have been seen about as often as Bill Gates at Linux World. (Trivia question of the week: Name the Massachusetts-based company that had the only commercial implementation of RAID 2, which used bit-striping for speed.) Additionally, combinations of RAID Levels 1 and 0 (usually written 1+0) and of Levels 0 and 1 (usually written 0+1) have also been popular. 1+0 is mirrored striping and 0+1 is striped mirroring (despite being combinations of the same levels, they are not the same thing).
Two other presumed forms of RAID, Levels 6 and 7, were proprietary solutions pushed by some vendors, and while in some cases they were technologically successful, they failed in the marketplace due to the essentially proprietary nature of their solutions.
RAID 7 was wholly proprietary to Storage Computer, a New Hampshire-based firm, which today is a shell of its former self.
RAID 6 looks a lot like RAID 5, except that parity is written to two drives rather than one. Because RAID 6 performs two different parity computations rather than one, it provides increased fault handling and can tolerate two concurrent device failures where the other RAID levels only support a single failure.
The extra calculating and writing to a second parity disk means that write operations for Level 6 are slower that for Level 5, but RAID 6 has not succeeded in the marketplace because it has no industry standard to which it can conform. This of course means that every implementation until now has been proprietary.
Now a number of industry players are banding together to standardize RAID 6, and there is some likelihood that it may actually become a proposed industry standard by year-end. If this actually happens - Intel is supporting this movement, so it may become a standard - RAID 6 may well become the RAID level of choice for read-intensive applications such as video-on-demand and other fixed content implementations. The industry players involved include: Adaptec, Fujitsu, HP, IBM, Intel, LSI Logic and Promise Technology.