RAID6 Technology Advances Disk Array Reliability

RAID5 has been popular because it provides combined performance from its member drives and reasonable protection against single disk failure. However, when RAID subsystems grow larger and its applications more complex, the RAID5 protection can be insufficient. In the event of a single drive failure, the occurrence of bad blocks on yet another member drive can render the affected data stripes unusable.

With RAID5, the recovery process following a drive failure is called "rebuild". Rebuild is initiated by either replacing the failed drive, or rebuilding to a hot spare that exists in the system. During rebuild, the array is vulnerable to a second drive failure, which is why the array is said to be critical at that time. RAID6 addresses and eliminates this vulnerability.

RAID6 improves on RAID5 and provides significantly higher redundancy in terms of its ability to withstand the simultaneous failure of up to two of its member drives, or the failure of an additional drive during rebuild initiated by a single drive failure. Put simply, in the unlikely event of a drive failure during rebuild, RAID6 is robust

enough to ensure continued access to data, RAID5 is not. This figure illustrates the additional parity blocks allocated to RAID6 redundancy:

RAID6 is similar to RAID5 in the way data is written to the array, but instead of just one, two parity blocks are available within each data stripe across the member drives. Each RAID6 array uses the equivalent of two member drives for storing parity data. The RAID6 algorithm computes two separate sets of parity data and distributes them to different member drives when writing to disks. In terms of capacity, RAID6 capacity is reduced from RAID5 by the equivalent of one member hard drive. This is a small price to pay for the added benefit of assured uptime.

One apparent disadvantage of RAID6 is that it consumes more processing power to calculate two sets of parity data at once during a write. In tests, we have seen a substantial reduction in array throughput when the RAID6 algorithms are implemented in the controller firmware without dedicated compute support from custom chips. Infortrend has recently introduced versions of their powerful self-contained RAID controllers with dedicated on-board ASIC 400 hardware support for RAID6, essentially negating the performance drop seen with RAID6 in software, while providing the important benefit of the ability to recover from multiple drive failures.