New Roads to Reliability

When it comes to disk drives, cars, suits, and nearly everything else: You get what you pay for. So while ATA disk drives have already earned a place in SANs, it's no secret that these inexpensive drives bring a tradeoff in reliability.

The higher failure rate of ATA offers little comfort to IT managers trying to do more with less. Storage buyers opening their arms to the cheaper variety add risk to their environments, while mission-critical applications demand more costly options.

But some new developments could ease the tradeoff between cost and reliability. Nobody's saying you should hold your breath waiting for ATA drives to get more dependable, or for windfall capex budgets that allow you to purchase more expensive drives. We're talking about developments such as Distance 4, Ice Cube, and the Oracle Resilient Low-Cost Storage Initiative. (See IBM R&D Aimed at Storage and Oracle Teams With Storage Vendors.)

These solutions take a holistic approach to drive reliability, making it possible to deploy more reliable systems with cheaper components.

Storage technologies that operate under the "sum is greater that its parts" principle are hardly new. RAID striping relies on this principle and new drive technologies advance it. For example, proprietary extensions to RAID 5 (sometimes referred to as RAID 6) achieve this goal by protecting against two drive failures instead of one.Distance 4 schemes aim to boost redundancy beyond two drives. RAIDn, developed by InoStor Corp. (Oslo: TAD) and available today, has this exact goal in mind (see InoStor Granted RAID Patent). Meanwhile, the brains at the IBM Corp. (NYSE: IBM) Almaden Research Center are cooking up two variations on Distance 4: one optimized for storage efficiency and the other for performance. The IBMers claim these implementations would yield a thousandfold improvement in reliability, but will likely take a couple years to materialize in products.

A more radical development at Almaden involves a three-dimensional stack of storage blocks, codenamed Ice Cube, an homage to its water-based cooling system and now known as CIB. Each self-contained "brick" houses an array, a processor, and network connectivity. A brick plugs into the stack and interconnects with the bricks it touches on all six sides, thereby scaling easily into petabytes of capacity.

The modular system also includes software that addresses availability. If one brick crashes, taking its array with it, the rest of the system picks up the slack. This idea flies in the face of an administrator's first instinct to replace failed components as quickly as possible. No need with CIB, say the project's engineers. The IBM system would keep on chugging and wait patiently for regularly scheduled maintenance, whether that comes in a week, month, or year.

A more rudimentary version of this concept already appears in a new initiative led by Oracle Corp. (Nasdaq: ORCL), of all vendors. The Oracle Resilient Low-Cost Storage Initiative, launched two weeks ago, calls for building a network of entry-level arrays behind Oracle databases. The "resilient" aspect of the implementation comes from RAID in the arrays and mirroring across them.

The goal is to create a storage grid of modular arrays that are low-cost and highly available, with unlimited scaleability. "The architecture does not depend on each array being reliable you just eliminate each array as a single point of failure, similar to clustering for servers," Oracle VP of systems technologies Juan Loaiza told a group of conferees at Oracle OpenWorld earlier this month.Loaiza says the most appropriate targets for this technique are the auxiliary databases behind production databases, as well as standby, reporting, testing, and disk-backup databases.

Oracle's IT department has dabbled with this concept, deploying four Apple Computer Inc. (Nasdaq: AAPL) Xserve arrays for the recovery area of a 4,000-user database. Administrators feared backup performance would suffer, but Loaiza reports little performance hit and $200,000 saved on new storage after five months.

Where does the savings come from? Loaiza claims the price/performance comparison between ATA and Fibre Channel is mostly a wash, or even favors ATA with certain types of sequential throughput. Also, enterprises can buy more arrays with the same amount of money, increasing the size of their storage grids.

"You get high availability using low-cost components because you build a system that can tolerate their failures," Loaiza says.

Besides Apple, Oracle has signed up Dell Inc. (Nasdaq: DELL), EMC Corp. (NYSE: EMC), EngenioInformation Technologies Inc., Hewlett-Packard Co. (NYSE: HPQ), MPC Computers, and Network Appliance Inc. (Nasdaq: NTAP). Now we'd like to see some customers, besides Oracle.— Brett Mendel, Senior Analyst, Byte and Switch Insider