Information Technology Reference
In-Depth Information
RAID Levels
An early paper on RAIDs, “A Case for Redundant Arrays of Inexpensive Disks
(RAID),” by David Patterson, Garth Gibson, and Randy Katz described a range of possi-
ble RAID organizations and named them RAID 0, RAID 1, RAID 2, RAID 3, RAID 4, and
RAID 5. Several of these RAID levels were intended to illustrate key concepts rather
than for real-world deployment.
Today, three of these variants are in wide use:
RAID 0|JBOD. RAID level 0 spreads data across multiple disk without re-
dundancy. Any disk failure results in data loss. For this reason, the term RAID is
somewhat misleading, and this organization is often referred to as JBOD—Just a
Bunch Of Disks.
RAID 1|Mirroring. RAID level 1 mirrors identical data to two disks.
RAID 5|Rotating Parity. RAID level 5 stripes data across G disks. G 1
of the disks in a stripe store G1 different blocks of data and the remaining disk
stores a parity block. The role of storing the parity block for different data blocks
is rotated among the disks to balance load.
Subsequent to the “Case for RAID” paper, new organizations emerged, and many
of them were named in the same spirit.
Some of these names have become fairly
standard.
RAID 6|Dual Redundancy. RAID level 6 is similar to RAID level 5, but
instead of one parity block per group, two redundant blocks are stored. These
blocks are generated using erasure codes such as Reed Solomon codes that
allow reconstruction of all of the original data as long as at most two disks fail.
RAID 10 and RAID 50|Nested RAID. RAID 10 and RAID 50 were origi-
nally called RAID 1+0 and RAID 5+0. They simply combine RAID 0 with RAID 1
or RAID 5. For example, a RAID 10 system mirrors pairs of disks for redundancy
(RAID 1), treats each pair of mirrored disks as a single reliable logical disk, and
then stripes data non-redundantly across these logical disks (RAID 0).
Many other RAID levels have been proposed. In some cases, these new “levels”
have more to do with marketing than technology (“Our company's RAID 99+ is much
better than your company's puny RAID 14”). In any event, we regard the particular
nomenclature used to describe exotic RAID organizations as relatively unimportant; our
discussion focuses on mirroring (RAID 1), rotating parity (RAID 5), and dual redundancy
(RAID 6). Other organizations can be analyzed using principles from these approaches.
Search WWH ::




Custom Search