There is a wide choice of hard drives from established players such as Hitachi, Maxtor, Seagate and Western Digital. For home or office use, a single disk is usually sufficient. But for enterprise networks, a cluster of several disks is often essential. The only problem is that these so-called RAID arrays are not exactly cheap.
Many motherboards have already built in RAID controllers, but not all of these are suitable for fileserver applications. Most of them only support the simple modes 0 and 1, and the combination of both called RAID 0+1 (striping & mirroring). While high capacity as well as outstanding performance can be accomplished by RAID 0, the risks of losing data due to a broken drive multiples by the amount of drives likewise. A happy medium between performance and data security can be achieved by a RAID 0+1, mirroring two RAID 0 arrays. The only downside is that the net capacity will be split in half.
If high capacity is first priority, the common option is a RAID 5, which distributes data to all array members and adds rotating parity information, too. The net capacity of the whole array is the sum of all drive's storage minus the capacity of one single drive. However, this solution is not as simple as it seems. Calculating the parity information as well as the real time reconstruction of data in case of a defective drive requires a lot of CPU performance. This forces the user to either buy a pretty expensive RAID controller with a built-in XOR unit, or to rely on a cheaper model that burdens the CPU with the XOR calculation.