SisSandra Disk Benchmark

Overall Score

Here, you can see that RAID 0 in general delivers the highest scores with the exception of the 512KB and 1024KB stripe sizes. You can also see that Software Stripe that was created through Windows 2000's Dynamic Disk system was almost as fast as hardware RAID 0. That's proof that the card itself doesn't do any processing.

Mirroring (RAID 1) is the worst performing RAID level. This actually makes sense because the system needs to write into two disks, instead of just one. And as expected, the performance of the spanned disks cannot be differentiated from that of a non-RAID single disk setup.

Buffered Read

Here, you can see that a RAID 0 array with a small stripe size (<16KB) generally gives the best performance.

Sequential Read

Ah... This is where the advantage of RAID 0 is really apparent. With striping, sequential reading performance was doubled for those with a small stripe size (<64KB). This means that the files used in the benchmark was quite evenly spread out between the two disks, effectively allowing the data to be read in parallel.

For bigger stripe sizes, the stripe blocks might actually be bigger than the files themselves, thus allowing more than a file to fit into a single stripe block. This means that when reading such files, the RAID 0 array can only achieve a transfer rate similar to that of a single disk. This shows in the performance hit in tests with stripe sizes of 128KB and above.

I actually expected the Mirror (RAID 1) array to perform much better. This is because during reads, it is possible for RAID 1 to achieve a sort of pseudo-parallel data transfer where each disk would serve different file requests. Even if the software isn't that intelligent, they could also show a marked decrease in seek time as the two disks can simultaneously seek different areas of the disk. However, the RAID 1 performed only as well as a normal non-RAID single disk setup. I can only conclude that the Promise FastTrak100 ATA RAID card does not implement such optimizations.