What would you do if data at your company was growing 100% to 150% annually? At the Fred Hutchinson Cancer Research Center in Seattle, the answer was a two-tier – and soon-to-be three-tier – storage architecture that uses hierarchical storage management software to keep the most-current data on the highest-performance platform and moves older data onto tape.
What would you do if data at your company was growing 100% to 150% annually? At the Fred Hutchinson Cancer Research Center in Seattle, the answer was a two-tier – and soon-to-be three-tier – storage architecture that uses hierarchical storage management software to keep the most-current data on the highest-performance platform and moves older data onto tape.
“Our research instruments are spewing more data that we have to collect and track, and it’s richer data, so it’s larger in size,” says Tim Hunt, research computing support manager at the center. “But we don’t want to keep purchasing expensive hard drives for old data – we want to be efficient.”
Hunt’s current architecture consists of a Sun e4800 server that acts as a portal to a Sun 6320-based storage-area network (SAN) and a Storage Technology 1700 tape library with 678 slots. The e4800 server runs IBM/Tivoli Storage Manager back-up software, and Space Manager HSM software that migrates data from the SAN drives to the tape system, based on how full the file systems are and the last time the data was accessed. The tape library is connected to the e4800 via direct-connect SCSI cables, while the SAN is connected via two Fibre Channel switches for redundancy.
“We can set a threshold for our file systems that when they’re 60% full, the HSM software kicks in and migrates the files that haven’t been used,” Hunt explains. His group can manipulate the size of different users’ file systems to accommodate their varying data access needs. “Some people need to look at their data a year after it’s been created, while others only look at it a month after it’s been created and then never again,” Hunt says. “We size our file systems so that data either stays on disk longer or is migrated off to tape to make room for more important data.”
Once data is moved to tape, users still can access it – they just experience a 20- to 30-second delay. There have been no complaints about this delay, but Hunt is evaluating the idea of adding a faster third tier of storage based on Serial Advanced Technology Attachment drives, which are faster than tape but slower than the more-expensive Fibre Channel-based drives in the Sun SAN.
Hunt also is mulling the possibility of a fourth tier, using slower but higher-capacity tape drives than his current StorageTek system. This would house very old data that would otherwise be moved off-site, creating complexities of tracking it down if it were ever needed.
So far, by adding just the second tier, Hunt has saved $100,000 in SAN drives that he would have had to purchase to keep data available to users.
Hunt can see the usefulness of being able to migrate data based not just on last access date but also on its value – an approach ILM proponents are touting. Currently, however, “I can’t distinguish what data is important and what’s not,” he says, “and I’m not sure I want to. I’ve got enough to worry about, with keeping up with data growth and making sure people can access it.”
At the same time, Hunt is glad these capabilities might be available in the future. “I’m glad people are talking about it because it’s a real issue for those of us who deal with a lot of data. We have a lot of gray hairs on our head,” he says.
