In I/O-bound applications, the differences between software-
and hardware-based RAID are less apparent. Clearly, if disk drives
are the bottleneck, whether the parity calculations are performed
in the main CPU or RAID co-processor will make little difference
in overall system performance. However, there are some situations
where hardware-based RAID may be advantageous.
You could see a significant improvement in mirrored drive (RAID
1) performance if you implement striping and mirroring
(RAID 0/1), not available on Windows NT or Netware software-based
RAID implementations. With RAID 0/1, not only could your
application experience improved read and write times due to
simultaneous multiple drive accesses, but also more consistent and
predictable performance due to the load balancing effect of RAID
0.
If your application is already I/O-bound, a failed drive in a
RAID 5 data set can have a paralyzing effect on system
performance. Hardware-based RAID solutions that support automatic
failed drive detection with hot spare replacement can
significantly reduce the amount of time your application is
running in degraded mode, because the application does not have to
wait until you physically replace the failed drive. Hardware-based
RAID solutions that allow you to set the priority
(low/medium/high) for array reconstruction, gives you control over
the tradeoffs you are willing to make between overall system
performance and availability.
Hardware-based RAID solutions can improve system boot time and
operating system performance by striping the operating system
files, a feature not supported on NT's software-based RAID
implementations.