Even when drives are RAID protected, it can take several hours to rebuild from a drive failure. Randy Arnott, RAID architect with startup RAIDCore, estimates that it would take 10,000 seconds, or 2.7 hours, to rebuild a single drive in an eight-drive 1TB RAID set, assuming rebuild rates of about 100MB/s.

Granted, "that's still better performance than most SCSI arrays," Arnott says. But considering that today's biggest SCSI drive (146GB) is less than half the size of today's largest ATA drives, you are limited in how big of a RAID set you can build--and consequently, how long it takes to recover from a failure.

There are also those that worry about a second disk failing before a RAID set has been rebuilt. That concern is behind Network Appliance's recent announcement of RAID-DP, which stands for dual parity. By adding a second "diagonal parity" disk drive into the RAID set, RAID-DP "lets us withstand a double disk failure and keep on running," says Mark Santora, NetApp senior VP of marketing. This in turn, "allows us to increase the reliability of currently inferior technology [i.e., ATA drives] and move them into the primary storage domain."

More likely than a double disk drive failure, says Keith Brown, NetApp director of technology, is a media error during a rebuild, "which is the equivalent of a double disk drive failure," he says. "And ATA drives produce more media errors, incidentally."

Of course, mirroring lets you withstand most double disk drive failure scenarios, Brown concedes. "But if you mirror your drives, you lose the cost advantage of ATA," he says. In contrast, RAID-DP adds no additional cost to the RAID set, and only minimal performance overhead, Brown claims. RAID-DP doubles the default size of a RAID set from six data and one parity drive, to 12 data and two parity drives. Also, Brown adds, by striping data across more disk drives, RAID-DP also helps boost performance.

RAID-DP will be available on arrays running NetApp's Data ONTAP 6.5 operating system.