By Dave Simpson
RAID 6 isn’t new. Vendors such as Hewlett-Packard, Hitachi Data Systems, IBM, and Network Appliance have been incorporating RAID-6 technology in their disk arrays for years (although they typically don’t refer to it as RAID 6). But there has been renewed interest in the technology, primarily due to the popularity of high-capacity Serial ATA (SATA) drives.
However, there are a number of misconceptions regarding RAID 6. For example, RAID 6 is usually defined as the ability to protect data in the event of two simultaneous drive failures. But since that event is extremely unlikely (regardless of drive type), it’s more accurate to say that RAID 6 protects against the failure of Drive B while failed Drive A is being rebuilt. And since the higher the capacity of a disk drive the longer the rebuild time, RAID-6 protection becomes most important in the case of SATA drives, which can store up to 500GB.
Another misconception is that RAID 6 has come to the forefront because of the reliability drawbacks of SATA relative to Fibre Channel and SCSI drives. Although relative reliability is a hotly debated issue, the high capacity of SATA drives is much more of an issue than their reliability specs.
Most RAID-6 implementations are based on disk striping with double distributed parity.
“The issue with RAID 6 is the capacity of the drives, not reliability,” says Manish Goel, vice president and general manager, data protection and retention solutions, at Network Appliance. “RAID 6 is most relevant in SATA environments because the drives go up to 500GB.”
Network Appliance’s RAID-6 implementation is called RAID-DP (double parity, also referred to as dual parity). Introduced in late 2003, RAID-DP is available on all of Network Appliance’s NearStore and FAS disk systems.
Goel also notes that RAID 6 protects against bit errors occurring during reconstruction of a drive (although bit-error rates on SATA drives are only slightly higher than on Fibre Channel drives).
Although RAID 6 is relatively new in the mainstream, Goel says that “most of [NetApp’s] customers that are using SATA drives are using RAID 6.”
NetApp’s RAID-DP implementation of RAID 6 differs from other vendors’ RAID-6 implementations. For one, all of NetApp’s RAID implementations are software-based. (RAID-DP is used in the Data ONTAP kernel.) In addition, NetApp uses a 14+2 (14 primary disks and two parity disks for RAID 6) scheme. (NetApp uses a 7+1 configuration in its RAID-4 implementations.)
A number of other disk subsystem vendors announced support for RAID 6 this year. For example, Nexsan Technologies’ entire line of disk subsystems (SATABeast, SATABoy, and SATA-Blade) is now available with controller-based RAID-6 protection.
Diamond Lauffin, Nexsan’s senior executive vice president, estimates that less than 10% of the company’s customers use RAID 6 today, but that “over the next year there will be substantial end-user adoption of RAID 6 as users understand the advantage it provides.”
In most cases, the RAID-6 functionality is available to users at no extra charge. However, it does require an extra drive, which can increase the overall cost of the RAID array. Lauffin notes that, even though SATA drives are inexpensive, the additional cost can prove to be a barrier in the application scenarios where SATA drives are most frequently used (e.g., secondary storage applications such as disk-to-disk backup, nearline storage, etc.).
RAID 6 + IP SANs
RAID 6 may find early acceptance in iSCSI-based IP SAN environments, where low-cost SATA drives have become the dominant drive type. A number of vendors that specialize in IP SANs and iSCSI recently introduced support for RAID 6.
For example, StoneFly Networks, a subsidiary of Dynamic Network Factory (DNF), last month announced support for RAID 6 in its iSC-1620, iSC-2420, OptiSAN, and ValueSAN IP SAN configurations. The company’s IP SANs with RAID-6 support start at $13,500 for a 2TB iSC-1620 configuration.
Similarly, start-up Nimbus Data Systems rolled out an iSCSI-based IP SAN last month that includes support for RAID 6 (see “Nimbus claims iSCSI ‘firsts,’ ” p. 12)
Controller-based RAID 6
A variety of controller manufacturers began implementing RAID 6 about a year ago. Examples include Adaptec, AMCC, Areca, Arena-MaxTronic, Infortrend, LSI Logic, Promise Technology, and Xyratex.
LSI Logic’s MegaRAID family of MegaRAID family of Serial Attached SCSI (SAS) adapters, which support both SAS and SATA drives, are “RAID 6-ready.”
Infortrend recently introduced support for RAID 6 via a firmware upgrade for its ASIC266-based RAID subsystems. Users can migrate a configured Logical Drive (LD) from RAID 5 to RAID 6. The RAID-6 firmware upgrade is available on most of Infortrend’s ASIC266-based EonStor controller subsystems, which include SAS, SATA-to-SCSI, and SATA-to-Fibre Channel configurations.
“Most end users with SATA-based subsystems have encountered bad blocks during drive rebuilds,” says Paul McLeod, technical sales and marketing manager at Infortrend, “so any user with concerns about reliability will be candidates for RAID 6.”
Some controller vendors use Intel’s X-Scale-based IOP33/IOP331 (PCI-Express/PCI-X) hardware accelerator engines to provide support for RAID 6. For example, Promise Technology’s recently introduced 16-port SuperTrak EX16300 and EX16350 SATA RAID controllers support RAID 6 via Intel’s chips. Promise also supports RAID 6 on its 8-port RAID controllers.
“Most end users aren’t aware of RAID 6 yet-although many resellers are hungry for it-but any user with a large number of spindles and high-capacity SATA drives will be interested in RAID 6,” says Suresh Panikar, vice president of worldwide channel sales and marketing at Promise.
Aside from the cost of an extra drive (for the additional double-parity calculations), the only penalty associated with RAID 6 is an inherent write-performance penalty. (RAID 6 does not exact a performance penalty for read operations.) In the early days of RAID 6, this penalty could be severe: a 30% to 40% (or worse) performance penalty versus RAID 5. However, most vendors have mitigated the write-performance penalty to the point where it is negligible.
For example, many vendors claim a performance penalty of only 2% to 4% vs. RAID 5. And even in cases where there is a 10% to 15% write-performance penalty, that would not be significant in the applications in which SATA-based arrays are typically used (e.g., nearline storage, disk-based backup, etc.). It should be noted the performance penalty associated with RAID 6 depends on a number of factors beyond a particular vendor’s implementation, including block size and whether the data stream is sequential or random.
In addition to the companies already mentioned, vendors such as Ciprico, JMR Electronics, and Tandberg Data also support RAID 6 on some of their disk arrays.