ZDNet UK

CBS INTERACTIVE UK BUSINESS SITES:
BNET UK
silicon.com
ZDNet.co.uk

You are here: ZDNet.co.uk > News > Hardware

Server platforms Toolkit

Picking apart RAID

Scott Lowe Tech Republic

Published: 13 Aug 2002 10:18 BST

Anyone who has worked with RAID has heard the term "parity". While most IT pros understand the general concept of parity, many would be hard-pressed to define exactly what it is or how to fix problems associated with it. Parity is a form of error correction commonly used in certain levels of RAID and works to reconstruct data on a drive that has failed in an array. In this article, I will focus on parity problems commonly associated with RAID levels 3, 4, 5, and 6. The remaining RAID levels either do not use parity or are not as commercially viable as these levels.

A lesson on RAID levels

The levels of RAID make use of physical disks in diverse ways. Each RAID level that supports error correction (parity) uses the capability in different ways as well. Table A explains these differences, as well as what can happen when a drive or drives in a RAID array fail.

Table A

Common name	Parity used	Description	Array's capacity	Data reliability	Minimum drives required	Failure condition
Disk striping	No	The data is broken down into blocks, and each block is written to a separate disk. Since the I/O load is spread across each disk, this RAID level performs efficiently both reading and writing data to and from the array.	Individual disk capacity multiplied by number of disks	Low	2	When one drive fails, the entire array is compromised.
Disk mirroring	No	All data is duplicated on both disks. This RAID level requires an even number of disks.	Number of disks divided by 2	Very high	2	When both drives fail, the data is lost.
Hamming Code ECC	No	There are two sets of drives: data drives and ECC drives. Each data word has its ECC data recorded on ECC disks.	Number of disks divided by 2	Very high	2	Because of its ECC encoding method, it is very inefficient and expensive. RAID 2 is not commercially accepted.
Parallel transfer disks with parity	Yes	Data sector is subdivided and distributed across all data disks. Redundant information is stored on a dedicated parity disk.	(N-1) disks	Very high	3	When more then one drive fails, the array is compromised.
Independent data disks with shared parity blocks	Yes	Each block is written onto a data disk. Parity for same rank blocks is generated on Writes, recorded on the parity disk, and checked on Reads.	(N-1) disks	Very high	3	When more then one drive fails, the array is compromised.
Independent access array without rotating parity	Yes	Data sectors are distributed as with disk striping. Redundant information is interspersed with user data.	(N-1) disks	Very high	3	When more then one drive fails, the array is compromised.
Independent Data disks with two independent distributed parity schemes	Yes	Like RAID 5, parity is striped across the disks in RAID 6. However, in RAID 6, parity is written twice.	(N-2) disks	Very high	3	Depends on implementation, although I know of no RAID 6 implementations.

In RAID 3 (Figure A), each file is broken up into blocks of identical size, which are then written to a disk in the array. The size of the block depends on the number of data disks in the array. With RAID 3, one disk is devoted to parity.

Figure A

Under RAID 4 (Figure B), an entire block of data is written to a disk before writing the next block to the next disk. This results in a file being written across multiple disks but not necessarily evenly. Like RAID 3, RAID 4 uses a separate parity disk.

Figure B

Like RAID 4, RAID 5 (Figure C) writes blocks of data to a disk before moving on, which means that one disk may store a larger chunk of data than another disk from the same file. Unlike RAID 4, however, RAID 5 stripes parity across the disks. To achieve its level of resiliency, RAID 5 requires the overhead equivalent of one of the disks in the array for parity. The more disks added to the array, the lower the percentage of overhead. For example, with three disks, one-third of the space is dedicated to parity. However, with six disks, only one-sixth is used.

Figure C

RAID 6 (Figure D) works almost identically to RAID 5. RAID 6 also stripes the parity across all of the disks in the array, but it is written twice, which allows for the failure of more than one disk. Unfortunately, it requires twice as much overhead as RAID 5.

Figure D