FASTORA DAS-315SA Disk Array
114
the files are also lost.
Logically, RAID 3 is RAID 0 with added fault tolerance.
A.3 RAID 1
This version of RAID has been around for years, known as “mirroring” or
“duplexing”. The exact same data exists on pairs of disk drives, and can be
read from either. This allows two simultaneous reads per mirrored pair, which
can double the read transaction rate. Write operations, which must be done to
both disks in the mirrored pair, offer no performance advantage, and may
actually reduce write performance well below that of a single disk.
Because of its high data reliability and ease of implementation, RAID 1 has
been used in storing critical data for years. Theoretically it is the most
expensive fault tolerant technique, due to having twice the disks required to
store the data, and does little to enhance performance.
A.4 RAID 2
Each disk in the array contributes one bit of the data transfer word; e.g., a
32-bit word requires 32 data disks. A solid-state memory Error Checking and
Correction (ECC) scheme called Hamming Code provides fault tolerance.
While Hamming Code requires several additional ECC disks (7 for 32 bits of
data, 8 for 64 bits of data), it is capable of identifying and correcting single disk
failures on the fly, so throughput is not affected by a disk failure.
Because all disks are involved in each data file access, the transfer rate is very
high while the transaction rate is that of a single disk. If spindle synchronization
is not used, average latency approaches a full disk revolution. RAID 2’s high
transfer rate is targeted at
very
large data record sizes, which it can efficiently
handle. It is seldom seen outside of the special application and supercomputer
