RAID technology -- Redundant Array of Independent Disks -- lets you establish varying degrees of data protection depending on the particular requirements of a given application. RAID levels 0, 1, 5 and 10 have been the most widely used, with RAID 5 (rotational parity) reigning supreme for fault tolerance because it allows data on a failed drive to be rebuilt without losing accessibility to stored information. RAID 6 (double parity) provides a higher level of fault tolerance by protecting data on two drives in the event of a failure.

In a RAID 5 array data is striped across all drives and parity information is distributed and stored across all the disks. If a drive fails the surviving array operates in degraded mode until the failed drive is replaced and its data is rebuilt from the parity information. But all data will be lost in the event of a second drive failure during a rebuild or a latent media defect that causes a read error during the rebuild. Today's increased hard disk capacities are causing longer rebuild times, which increases the likelihood that a second drive will fail during rebuild.

RAID 6 eliminates that risk. In a RAID 6-enabled system a second set of parity is calculated, written and distributed across all the drives. This second parity calculation provides significantly more fault tolerance because two drives can fail without resulting in data loss (see graphic).

But the additional calculations required with RAID 6 adversely affect write performance. Performance benchmarks show a RAID controller can suffer more than a 30 percent drop in overall write performance compared with a RAID 5 implementation. RAID 5 and RAID 6 read performance are comparable.

RAID suppliers implement their designs differently, so it is important to find controllers that minimize the RAID 6 write penalty. Look for controllers that make double parity calculations simultaneously and that use dedicated, silicon-based stripe handlers to reduce the write penalty radically.

Capacity considerations

RAID 5 implementations require a minimum of three drives and have the storage capacity of N-1 drives because the equivalent capacity of one drive is exclusively dedicated to holding parity data. For example in a four-drive, 200GB-per-drive array, the available storage capacity is 600GB out of a total of 800GB.

RAID 6 implementations require a minimum of four drives and have the storage capacity of N-2 drives because the equivalent capacity of two drives is exclusively dedicated to holding parity data. The available storage capacity is 400GB of 800GB.

While RAID 6 can be used with as few as four drives, RAID 10 is a higher-performing configuration that tolerates most two-drive failures in a four-drive array. RAID 10 mirrors and stripes data to maximize redundancy and performance. A RAID mirror does not require a read-modify-write operation as RAID 5 and RAID 6 arrays do. RAID 6 is recommended over RAID 10 for implementations with more than four drives.

In summary, RAID 6 provides higher levels of data protection, data availability and fault tolerance than RAID 5, but it comes at a cost. RAID 6 requires the equivalent capacity of two drives in the array to be dedicated to storing parity information, and most RAID 6 systems carry a heavy write-performance burden because of the additional parity calculation and the additional memory interruptions. Simultaneous parity calculations can mitigate these performance impediments.

Eischen is senior marketing manager for AMCC's storage division. He can be reached at ceischen@amcc.com.