Parity pioneer combats data corruption

Two of the latest arrivals to my lab — a couple of 1TB Barracuda ES2 (Enterprise SATA 2) drives from Seagate — got me to thinking about a growing problem in the enterprise, namely, that our voracity for capacity could quickly spell disaster if we fail to compensate for drive failure rates.

[ Mario Apicella’s column is now a blog! Get the latest storage news from the Storage Adviser blog. ]

But first, my new toys, those 1TB ES2 drives, which quickly found a home in my 16-slot enclosure.

To put that 1TB capacity into perspective, consider this: Filling up all 16 bays with similar drives would give me enough capacity to store most if not all of the Library of Congress in digital format in three rack units, just over 5 inches.

Of course, you could obtain that same capacity with 16 of the Barracuda ES2’s desktop cousin, the Barracuda 7200.11, as both offer 1TB models. But take heed: The 7200.11’s BER (bit error rate) is an order of magnitude worse than that of the ES2, which boasts one permanent read error for every quadrillion bits read.

What that means in practical terms is that, with the 7200.11 drives, a read error is statistically certain for every 12TB read (I am rounding up for clarity). In essence, if I had my 16-slot array filled with these drives in RAID 5 configuration, a read error during rebuild — and the consequent loss of data — would be very likely.

With the more reliable ES2 drives in place, the error would be likely after the drives read 120TB, thereby fending off data loss — but not for long, given how fast drive capacities are growing.

To address this problem, many in the industry are beginning to adopt dual-parity, aka RAID 6. Some vendors, however, remain unconvinced that dual-parity is enough.

“RAID 6 treats the symptoms — failed reconstructions — but not the cause of the problem, which is media error,” objects Larry Jones, vice president of marketing at Panasas.

Tiered parity, a new technique developed by Panasas, takes care of those media errors, according to Jones. The technology will be included in the next version of Panasas’ OS, Active Scale 3.2, expected to ship by year’s end.

As the name implies, tiered parity addresses media errors by defining tiers of reliability checks, including vertical, horizontal, and network parity.

Horizontal parity is nothing new. Panasas has been using this unique multi-controller, data-aware RAID technique to deliver incredibly fast rebuilds for years. Horizontal parity will save the bacon faster in the event of a drive failure than most any other RAID approach.

Vertical parity, essentially RAID inside a single drive, however, is not only new but nothing less than revolutionary. Imagine segmenting each drive in clusters of sectors and assigning a parity function to specific sectors inside each cluster. When a read error occurs, the vertical parity algorithms rebuild the compromised data from the parity sectors, avoiding both a long rebuild and possible data loss.

I wish I knew more of the implementation details of vertical parity — for example, how many sectors are in a cluster and how many are set aside for parity? But it’s understandable that Panasas is revealing as little as possible about its jewel while the patent is pending.

“The parity data amounts to about 10 percent of the drive,” Jones explains, adding that vertical parity is designed to adjust to future drives larger than 1TB.

As for the third leg of the tiered parity paradigm, if you can wrap your mind around horizontal parity, network parity should be easy to understand: Add a parity computation to disk data so that possible bit losses along the data path can be easily spotted.

Panasas’ parity trifecta has the potential to change your approach to data protection significantly. But don’t expect to see vertical parity from other vendors soon. According to Jones, the company intends to keep its new technology proprietary, at least for now.

Join me on The Storage Network with questions or comments.