Technical information

62
Overview of RAID
What is RAID?
One of the most popular server requirements is RAID, an acronym for Redundant Array of Independent (or
Inexpensive) Disks. The key concept in RAID is this: a single disk failure does not bring your system down,
because you can recreate the lost data from parity information stored on the other disks. Today RAID
addresses three needs:
Fault Tolerance:
today's applications typically require 100% up time.
Scalability:
the ever-increasing requirement for more resilient storage capacity.
Performance:
more simultaneous reads/writes, or faster data transfer.
The RAID Levels:
In 1987, three University of California researchers outlined the original RAID classifications. Their papers
sparked interest in disk arrays by end-users and storage vendors, and untangled a wide variety of thorny
design issues. The RAID levels they outlined are as follows:
RAID Functional Description Table
RAID 0
Called striping. The data is broken up into chunks as it is written across all the drives. This
provides the highest performance but no redundancy is provided. Strictly speaking the
absence of redundancy in a striped array makes the term RAID a misnomer.
RAID 1
Called mirroring. The data is written redundantly to pairs of drives and can be read
independently from each drive. This is fast and provides full redundancy but twice the needed
disk capacity is required.
RAID 2
The data is bit-interleaved across groups of drives with some of the drives storing error
correction codes. This provides full redundancy and more error correction capability, but is
often slow due to the hardware overhead.
RAID 3
The data is byte-interleaved across groups of drives with only one drive dedicated to storing
parity. This provides full redundancy and high transfer rates when large blocks of data are
transmitted. This is best for high bandwidth applications.
RAID 4
The data is broken up into chunks and written across the drives with one drive dedicated to
parity. This provides full redundancy but the single parity drive causes a bottleneck when the
parity must be updated.
RAID 5
The data is broken up into chunks and written across the drives. The parity for the stripes of
data is also spread across all the drives so no one drive is dedicated to parity. The operating
system sees the group of physical disks as one disk drive. This is the best for transaction
processing, where many small I/Os are required.
There are other types of RAID such as RAID 6, RAID 7, RAID 1+0 and RAID 3+5. These are hybrids of the
original RAID levels described above. e.g.
RAID 10
Striping of two RAID 1 arrays.
RAID 30
Striping of two RAID 3 arrays.
RAID 50
Striping of two RAID 5 arrays.