IBM Midrange System DS4000 Series Hardware Manual page 125

RAID 6: High availability with additional fault tolerance
RAID 6 (see Figure 4-4) is a RAID level that employs n+2 drives, which can survive the failure
of any two drives. RAID 6 stripes blocks of data and parity across an array of drives and it
calculates two sets of information for each block of data (p+q). For the purposes of RAID 6
p+q, they can be used to generate up to two missing values from a set of data. The key to this
method is the q, which is a codeword based upon Reed-Solomon error correction. As such, q
is more like a CRC than parity. Based upon principles of set theory and linear algebra,
Reed-Solomon codes are well-known codes that are also maximum distance separable
(MDS).
The calculation of q is complex. In the case of the DS5000 storage subsystem, this
calculation is made by the hardware and thus more performant than the software-based
implementation found in other storage systems.
By storing two sets of distributed parities, RAID 6 is designed to tolerate two simultaneous
disk failures. This is a good implementation for environments using SATA disks.
Due to the added impact of more parity calculations, in terms of writing data, RAID 6 is slower
than RAID 5, but might be faster in random reads thanks to the spreading of data over one
more disks.
Disk 1
A0
A1
A2
Q4
P4-parity
...
Figure 4-4 RAID 6
RAID 6 is striping with Dual Rotational Parity.
Logical Drive
Block 0
Block 1
Host View
Block 2
Block 3
Block 4
Block 5
etc.
Disk 3
Disk 2
B0 C0
B1 Q2
Q3
P2-parity
C2
P3-parity
C3
B3
...
...
RAIDset
Chapter 4. IBM System Storage DS planning and configuration
Disk 4 Disk 5
Q1
P0-parity
P1-parity
E1
E2
D1
Q0
D2
E3
D3
...
...
107