DAS-208SA Disk Array User Guide
78
both the old data file (to be overwritten) and its corresponding parity file
must be read;
a new parity file is generated from the old data file, the parity file, and
the new data file;
the new data file and the new parity file are written.
A read error requires that all disks be used to recreate the failed data file,
eliminating the multiple concurrent transfer potential.
A.7 RAID 5
To improve the poor write and read error performance of RAID 4, RAID 5
makes one simple (in concept) change which has major performance
ramifications: distributed parity files. Instead of having all parity files located on
a single disk, the parity files are allocated evenly to all disks, always with the
corresponding data files on other disks.
This change allows multiple writes to take place as long as they don’t “collide”
at a particular disk; e.g., in an eight disk RAID 5 environment, as many as four
concurrent writes could occur. Each write still involves two disk revolutions.
Read transaction performance is excellent, with as many concurrent reads
possible as there are disks. The aggregate read transfer rate is also excellent,
but each individual read is transferred at a single disk’s rate.
