Disk
Striping with Hamming Code for Error Protection
None
Yes
(1 Drive Failure)
RAID
3
Disk
Striping with Dedicated Parity Drive Parallel Disk I/O
Yes
(1Drive
Failure)
RAID
4
Disk
Striping with Dedicated Parity Drive; Non- synchronized
Disks Required
Parallel
Disk I/O
Yes
(1Drive Failure)
RAID
5
Disk
Striping with Distributed Parity
Parallel
Disk I/O (not as Fast as RAID 0)
Yes
Redundant Arrays of Independent Disks, or RAID, is a rapidly
expanding storage technology which promises a major improvement
in the way on-line data is stored in computers.
RAID Level Definitions Those investing in storage will need to consider low cost per
Mbyte, high input/output I/O, and high data reliability in order
to obtain a balance to suit their needs.
RAID 0
RAID 0 - Disk Striping Disk striping writes data across all disks concurrently rather
than on one disk at a time. Although termed RAID 0, it is not
a true implementation of RAID because there is no facility for
redundancy. Therefore, in the event of a disk failure, data is
lost.
In
the disk array subsystem, data chunk 0 is written to disk 0 ,
chunk 1 is written to disk 1 and so on. When the last disk is
reached and written, the array proceeds to store data on the next
level of the first disk.
Disk striping is fast as data can be transferred to multiple disks
simultaneously: chunk 0 is still being written to disk 0 while
chunk 1 is being written to disk 1. Furthermore, reads and writes
can overlap.
An example of a typical usage for RAID 0 could be: Data from the
field comes into the central processing location on tape where
it is instantly processed. Redundancy is not a requirement as
the tape can be relocated.
Summary: RAID 0 offers the highest performance without redundancy.
Some industries that RAID 0 is particularly suited to are: meteorology,
geophysical exploration, oil and gas industries, video/graphics.
RAID 1 - Disk Mirroring Disk mirroring protects against disk failure by keeping
two copies of data stored on separate disks or arrays. Though
simple and easy to implement, installing two sets of disks effectively
doubles the investment required for a single, non-redundant drive.
If at any time either disk fails, the remaining disk can provide
all of the data needed, preventing downtime.
Two copies of the data also ensure that there is no degradation
in performance, as accesses are immediately routed to the working
disk. In the event of failure, copying from the operational disk
to the replacement disk is very fast, which reduces the risk of
a second failure.
RAID 1
RAID 1 not
only provides protection, it can also improve performance. For
example, if multiple requests for the same data are made, demand
can be distributed between two disk copies therefore increasing
response time for data access.
Summary: RAID 1 is the most secure of any of the RAID levels and
is exceptionally fault-tolerant. Examples of industries that would
use this level are those who cannot afford downtime: banks, insurance
companies, stock markets, airline systems.
RAID
3
RAID 3 - Parallel Data Access In RAID 3 data is distributed to
a striped array and a disk is added to store redundant information.
The array consists of three disks for the data and one parity
disk for the redundancy. In the event of a disk failure, data
can be mathematically reconstructed from the remaining disks in
the array.
Synchronization enables striped data to be read and written
as quickly as possible. However, when multiple writes are involved,
performance is reduced because the parity drive has to be accessed
for every single write, which may create a bottleneck at the parity
drive. Consideration should also be given to impacts on performance
as disk rotation must be synchronized before data can be accessed.
Summary: RAID 3 is ideal for intensive high-speed, long data transfer
applications such as: video, CAD/CAM, graphic applications, scientific
modelling.
RAID 5 - Independent Access Arrays In RAID 5, the redundancy offered in RAID 3 by a single parity
disk, is distributed across all the disks in the array. Data and
relative parity are never stored on the same disk.
One user may be writing a chunk to disk 0 and the corresponding
parity to disk 3, another user may be writing to chunk 4 of disk
1 and updating parity on disk 2. There is a clear dividend in
terms of performance and the speed of transactions.
RAID 5
During disk writes, RAID 5 cannot produce a write performance
comparable to that of straight disk striping because other operations
have to be undertaken to make and store parity codes. The I/O
performance of the array depends very much on the relative levels
of reads and writes requested.
When a stripe is modified, unmodified portions must also be read
to re-generate the parity for the entire stripe. Once the parity
has been generated, the modified data and parity information must
be written to disk. This is commonly know as Read/Modify/Write
strategy.
It reflects that, though RAID 5 is superior to RAID 0 because
it offers redundancy, it is not able to perform as well as RAID
0 in terms of write performance. Because RAID 5 has distributed
parity, two reads and two writes must be performed for every write
operation. However, the write penalty can be overcome by the use
of write caching which allows write data to be stored in the memory
prior to writing to the disk, so freeing the host processor for
other tasks.
Summary: RAID 5 is ideal for organizations running databases and
other transaction-based applications such as: banks, airline and
railway reservation systems, government departments, utilities
and telecommunications.
