Children's Hospital trims backup with disk-to-disk

31.10.2005
Children's Hospital in Boston was dealing with a nightmare of a backup window earlier this year that at times lasted from 4 p.m. to 10 a.m. the next day and often entailed failed tape drives and shutdowns. On top of at least two drive failures each month, the sheer speeds of data transfers were in the 30MB/sec. to 40MB/sec. range, which guaranteed that the backup window for some 300 servers would continue to grow.

"The backup windows just kept growing and growing, and there were so many problems at night that backups were being missed," said Paul Scheib, director of operations and chief information security officer at Children's Hospital. "You start seeing these slower speeds but you really don't know what it is. It becomes like an Easter egg hunt where you're trying to figure out what's limiting the rate to the drives. Is it the drives? Is it the servers? Is it the network?

The hospital launched a pilot program of a virtual tape library (VTL) technology as an intermediary between its Fibre Channel storage-area network (SAN) and its two tape libraries, but that had problems of its down, according to Scheib. "We had a lot of trouble getting it configured correctly," he said.

Instead of sticking with a VTL, which required integration with his Fibre Channel SAN, Scheib said that in June he chose another disk-to-disk backup technology that removed Fibre Channel from the equation and put server backups on the LAN.

By using DD460 disk arrays from Data Domain Inc. in Palo Alto, Calif., Schieb said, he was able to replace one of his tape libraries with three disk arrays that cost about the same amount: US$75,000 each. The disk arrays doubled data transfer rates, removing three hours from the hospital's lengthy daily backup windows, reducing backup failures by half and speeding up restores from hours or days to minutes.

The disk-based backup technology also allows the hospital to keep about three months of backup data online, nearly eliminating off-site tape recalls for data restores. Time to respond to urgent restore requests has been reduced from four hours to a few minutes, Scheib said, adding that he was also able to eliminate one of his Hewlett-Packard Co. tape libraries.