All the were performed on the test system with WD360GD 10k RPM hard drives. The default stripe size was 16 kB, and the default NTFS cluster size was 512 bytes. The Silicon Image driver was version 1.0.0.40, and the FastTrak driver was version 1.2.70.14.62x.

Test A was a two disk RAID-0 on the Silicon Image Sil3112 RAID controller (embedded in A7N8X deluxe motherboard).

Test B was the same as Test A, except the RAID-0 stripe size was 4 kB, and the NTFS partition was formatted with a 4 kB cluster size.

Test C was a four disk RAID-0 on the FastTrak SX4 S150 RAID controller.

Test D was a two disk RAID-0 on the FastTrak SX4 S150 RAID controller.

Test A and B test the theory that equating the stripe size with the NTFS cluster size provides a performance improvement. Matching the two resulted in a five percent increase in random write performance, but about 10 percent drop in sequential and buffered write performance. Read performance was not significantly different. It is likely that the buffered and sequential write performance increases with stripe size and the cluster size choice does not have much effect on performance.

The second two tests show the performance of the identical test system, with the drives attached to the FastTrak SX4 S150 controller. As you can see, the performance is abysmal compared to the Silicon Image controller. Even the four disk RAID-0 has terrible performance, not reaching half the performance of the two disk RAID-0 with the other controller. The benchmarks do not indicate a cause, which could include a poor SX4 Windows XP driver, slower controller, PCI bus limitation, or some system setting or configuration.