Databases Reference
In-Depth Information
RAID 3 is yet another variation that uses a finer granularity of data I/O than other
RAID models, this time at the byte level. Bytes are written to disks in parallel (just like
with RAID 2 it was bits written in parallel), and in order to do reasonable I/O on data
(blocks) the disks need to be kept perfectly synchronized. Like RAID 2, this leads to
very poor concurrency for multi-user access on the storage array since all disks within
the array need to service a single request at the same time (working together). A disk is
usually dedicated from within the array to serve as a parity disk.
Figure 13.9
RAID 4.
13.7.5 RAID 4
RAID 4 stripes blocks of data across the storage array, and also introduces the use of a
dedicated disk to store parity bits (Figure 13.9). This is similar in concept to RAID 3
except that the I/O to the disks is performed at the block level, not at the byte level.
This is far more conducive to multi-user access.
13.7.6 RAID 5 and RAID 6
RAID 5 (Figure 13.10) is a variation of RAID 4 where the parity blocks are distrib-
uted across disks, instead of being stored on a dedicated parity disk. Most RAID 5
arrays include hardware support for the parity calculations. In general the combina-
tion of striping (parallelism) and parity bits makes RAID 5 one of the most popular
RAID schemes for database systems.
