"SATA drives aren't designed for high performance to begin with," said Brad O'Neill, senior analyst with the Taneja Group. "Wherever users, especially in cost-sensitive smaller organizations or departments and Tier-2 storage in larger organizations, are relying on SATA disks for business operations, it makes more sense to default on the side of better protection in exchange for a minor performance hit."
"The biggest benefit of dual parity is not protection against two simultaneous disk failures, which is a relatively unlikely scenario," said Tom Treadway, chief technology officer (CTO) of the storage software group at Adaptec Inc. "The real protection is against bit failure during rebuild mode."
Users weigh in
"We've had no issues with performance so far," said Andy Tran, senior executive vice president and CTO for Pacific Title and Art Studio, a post-production studio in Hollywood, Calif. Pacific Title is using half a petabyte (PB) of DataDirect Networks' 9550 SATA arrays, which get over the performance hump of calculating two-parity variables by packing extra hardware horsepower into the box in the form of 28 dual-core specialized ASICs surrounding a central cache.
SATA was attractive for storage during film processing because of its high capacity and low cost, Tran said, but the extra protection of RAID 6 is also important. In the course of making a film, Pacific Title creates huge volumes -- up to 30 terabytes (TB) -- out of groups of equally massive LUNs. "In my business we have to make sure that there's no corruption -- any LUN we lose, we lose the entire volume and could lose an entire movie," he said.
Still, Tran said, he has only deployed RAID 6 on one array so far and is working to become more familiar with it before implementing it further. He expects that in the long run he will depend a lot more on RAID 6 but said it's not for everyone -- especially in bigger arrays where devoting two drives per RAID group to parity adds up to higher disk costs quickly.
"I don't think too many [vendors] will provide RAID 6, not as widely as RAID 5," Tran said. "[Users] are also worried about capacity."
New approaches to the RAID 6 performance problem
For users considering RAID 6, some vendors are attempting to ease performance woes through more efficient parity-calculation algorithms, which can reduce the number of disk I/Os required in the process.
In a RAID-5 system, one disk per RAID group is used for calculating parity using an algorithm called exclusive, or in which the processor solves an equation for one value, used to replace missing bits in the event of a drive failure. RAID 6 requires that equation to be solved for two variables, meaning bits in traditional exclusive or processing schemes must be accessed twice. Thus, under the traditional algorithms, solving for two variables with six disks in a RAID group would mean 12 I/Os.
However, vendors offering RAID 6 schemes today, such as Overland Storage Inc., Adaptec and Nexsan Technologies, are instead using the Reed-Solomon algorithm, which allows dual parity to be calculated using one I/O per disk in the RAID group -- so six disks would require a minimum of six I/Os.
Other vendors, most notably Network Appliance Inc. (NetApp), are changing the way storage controllers within their boxes are reading data on disk to reduce RAID 6 I/O loads. For its part, NetApp is using its Write Anywhere File System (WAFL) to place dual-parity data on disk platters in compact stripes so that it can be more easily accessed by the controller during RAID calculation.
Overland has jumped into the RAID 6 space with a feature called DriveAlive, a kind of hot-swapping feature that staves off full-rebuild mode when one of the disk blades in the array chassis is removed for routine maintenance, such as replacing a drive platter. DriveAlive reduces performance issues in rebuilding mode by reducing the cases in which the array goes into rebuild mode at all.
Beyond SATA, RAID 6 may be limited
NetApp said it remains sanguine about the possibilities of RAID 6 as a data protection scheme for Fibre Channel (FC) and serial attached SCSI (SAS) arrays, but whether or not users of more expensive, higher reliability disk drives will ever see a need to stray from single parity is very much in doubt -- something even other disk vendors have said.
"RAID 6 is going to be a checklist item for some customers," said Jerry Hoetger, director of product management for Xyratex. "But I don't really see much interest for it with higher reliability drives, like SAS and Fibre Channel."
More and more users are packing bigger workloads onto less-expensive drives, but there are still cases where performance concerns will win out, according to O'Neill. "Users have to make workload-specific, sometimes application-specific, decisions" about what form of parity protection to use, he said.