Chicago Merc Ditches Tape Backups For Disks

The Chicago Mercantile Exchange Inc. has experienced exponential growth in its IT infrastructure as electronic trading volume for its futures and options contracts soars. In the past two-and-a-half years, its storage area network has grown from 4 terabytes to more than 180 terabytes, the number of servers has grown from 500 to more than 2,000 -- half of them Linux -- and its expanded from one data center to three.

That's raised a fundamental question -- where to store all that data? Chicago Merc's infrastructure is made up of four tiers of storage: Tier 1 for critical application data and databases that require high performance, such as trading systems; Tier 2 for noncritical production databases; Tier 3 for long-term archiving, such as that required for regulatory retention; and Tier 4, a tape-based system for backup and recovery.

The sore point was Tier 4. The Chicago Merc had two StorageTek tape silos that were controlled by mainframe-based software. Tape backups were slow and resulted in a high rate of backup and restore failures. Also, there were only two silos supporting three data centers, so an additional silo would need to be purchased.

The Merc hit upon the idea of adding a new tier 4, with disks replacing tapes as the medium for backup and restores, and bumping tape to tier 5, where it would be used only for off-site storage and disaster recovery.

The exchange evaluated a number of standard ATA-based disk products for tier 4. All suffered from one fundamental defect: ATA drives are built to have a "duty cycle" -- the percentage of time that a disk is spinning -- of between 25% and 50%. In the products that the Merc was investigating, the drives were spinning 100% of the time, meaning they were running at more than twice their duty cycle, and as a result had a higher rate of failure "What [the vendors] are doing is installing drives that have a 25% to 50% duty cycle in hardware where they're spinning 100%," says Craig Taylor, the Chicago Merc's associate director of open systems.

The Merc turned to an alternative product, Revolution 200T, from Copan Systems. Revolution 200T employs a radically different approach: Massive Array of Idle Disks, or MAID. In the MAID setup, a disk only spins when a piece of data that resides on it is requested; the rest of the time it sits idle. Because the disks are idle most of the time, there's less chance they'll fail. "I've implemented an architecture with ATA drives that spins them down when not in use," says Taylor. "The access isn't real-time, but it's backup data, so a 15-second delay is no problem."

In addition to greater reliability, the MAID system packs more storage into a smaller footprint; a 10-square-foot MAID system can hold up to 224 terabytes, whereas a 15-square-foot standard ATA system can only hold 56 terabytes. The MAID system works easily with the Merc's backup software, Symantec Corp.'s NetBackup; to NetBackup, the MAID system appears as a "virtual tape" library. "NetBackup has no idea that these aren't tapes," says Taylor.

In the year since implementing Revolution 200T, the Chicago Merc has expanded its total MAID storage to 340 terabytes, with ample capacity for growth. The comparisons between the MAID system and the Merc's old tape backup system are stark: Backups and restores are up to three times faster, and no backups have failed compared with a continuous series of failures on tape.

Taylor would like to extend the MAID concept to other tiers in the Merc's storage infrastructure, such as Tier 3, which currently employs EMC's Centera storage technology. He says, "I'm putting long-term data on [Centera] boxes with spinning disks inside. That's not a good solution for long-term storage."