For the linear scaling with I/O test, we used the RAIDCore controller, because it handles most of the functionality with software, which puts a heavier strain on the system processor (or processors). This is a disadvantage with multifunction servers that simultaneously assume data, print, email, database and web services. In that case, heavy access to the memory subsystem taxes the system processors unnecessarily. But this is not the case with dedicated servers: if a computer is only used for file serving services, the processor load can of course get heavier.
We assumed just such a case for this project, since we were mainly concerned with determining which level of increased performance could result from a larger number of hard drives in the ideal scenario. With two 2.8 GHz Xeon processors, the CPU load is guaranteed not to pose a problem.
As expected, the test arrays with three to eight drives scale extremely linearly, at least in terms of I/O performance. For database, web or file server applications, using several hard drives thus quickly pays off. But if you're going to use more than six drives, we always recommend integrating a spare drive as well that automatically takes over in case of a fault.
If you're mainly interested in high data transfer speeds, though, fewer drives will serve you just as well. Our RAID 5 array reached nearly 220 MB/s with five drives. Eight drives yielded only a slightly higher value, at around 237 MB/s.
If those performance values still aren't good enough for you, you might consider a configuration consisting of two PCI-X controllers. You can often tweak a clear performance increase out of them.