Finally, I again braved the long RAID rebuild times and created RAID 6 and 10 arrays. RAID 6 is commonly described as an extension of RAID 5 that allows for additional fault tolerance by using a second independent distributed parity scheme. RAID 10 is a striped array using RAID 1 arrays as segments and yields the fault tolerance of RAID 1, but with higher performance.

Thecus' manual says that both RAID 6 and 10 arras can have two drives go south and still keep running. As I mentioned earlier, I actually tried this while running an iozone test on a RAID 6 array and can confirm that the test kept running as I pulled and then reinserted two drives.

For those of you wondering about the relative performance of RAID 5, 6 and 10, just feast your eyes on Figures 26 and 27.

Figure 26: Thecus N5200 Write performance comparison - JBOD, RAID 5, 6, 10 - 1 Gbps (click image to enlarge)

It's a bit hard to tell given the scale of the plot and the natural variation in performance, but it looks like RAID 6 results in a performance hit that's close to what you get while a RAID 5 array is rebuilding. And it looks like RAID 10 is the way to go for the best combination of performance and data robustness.

Figure 27: Thecus N5200 Read performance comparison - JBOD, RAID 5, 6, 10 - 1 Gbps (click image to enlarge)