Qualcomm

LAS VEGAS -- Making sprawling data centers more efficient to manage across tiers is a major challenge for any large firm. To do it while saving money is an even greater challenge, although not impossible.

Admins from wireless chip manufacturer Qualcomm and Philips Lighting Electronics North America detailed how they pulled off their storage makeovers at a conference here. For Qualcomm, it was a matter of getting a greater handle on utilization and streamlining storage into one pool while Philips went from a monolithic to a modular SAN and embraced utility pricing.

As Qualcomm's storage manager, Paul Ferraro is responsible for managing more than 2 Pbytes of NAS data, 600 Tbytes on SANs, 1,500 storage ports, 4,500 servers, and 15,000 desktops.

Qualcomm is mostly a Network Appliance NAS shop and has SAN gear from EMC, Hitachi Data Systems, and 3Par. Storage resides in five data centers spread over two buildings. "We've got storage arrays scattered all over the place," Ferraro says.

In September, the San Diego-based firm started a consolidation project to streamline its data center, increase storage utilization, and reclaim data center space. Ferraro says that project will be completed next week.So far, Qualcomm has increased host utilization from 61 percent to 70 percent, gained about 10 to 12 tiles of data center space, and reduced its arrays from 33 to 15. Now its storage is in one pool so it can be reprovisioned more effectively across tiers. Ferraro wouldn't reveal the cost savings, but he calls them "huge."

The consolidation involved migrating 50 percent of Qualcomm's storage while staying online. Ferraro says he managed the project with Symantec's Veritas Volume Manager for host volume management, Monosphere Storage Horizon, for capacity planning, and HDS Tuning Manager, for performance analysis. "And of course we used spreadsheets, still."

Ferraro also took steps to curb data growth by making users pay for their storage utilization, and only giving them more if their utilization rate is high. The Monosphere capacity planning software helps him determine how much storage each department is using, and how much capacity is unused. (See Capacity Planner Grabs More Cash.)

"If they have stuff they're not using and they ask for more, we tell them no," he says. "We show them reports -- here's the servers we have, here's the utilization. When they ask for more terabytes we look at the reports and say, 'You have 20 terabytes you're not using, where do you want to take it from?' It's painful for them because they're used to asking us for it and we give it to them."

"Sometimes they're good. They're utilizing 75 percent, so we give them more."Another way to keep users from asking for too much capacity is to charge them enough so they will think twice about asking for more."One of the things we did that was helpful was set up a price for storage," he says. "For example, $30 a gig for tier one, $20 for tier two, and so on. Then we figure out a price that's higher than what we pay. If we charge them the acquisition price, they'll say, 'Great, give me more.'"

While on a smaller scale, Philips Lighting systems and storage administrator Erik Windischman set out to solve many of the same problems as Qualcomm.

Windischman says when Philips' lease for its Hitachi Lightning system leased through HP expired last year, his company decided to move away from a monolithic SAN to a tiered approach. Philips also decided against leasing for HP's utility pricing model.

"Costs for new equipment were less than leasing," Windischman says. "And our financial people decided utility pricing would be best. You agree to set a price up, build in some overhead in capacity, but only pay for that storage when you allocate it."

Philips has about 20 Tbytes on its SAN now and knows how much it would cost to expand.Instead of a high-end Hitachi system, Philips acquired a mid-range Hitachi TagmaStore for its mission critical data and added a lower-cost FATA drive array from HP for tier two storage. (See Hitachi Plans Midrange Rollout, HP, Sun Upgrade Hitachi, and A New Landscape for Disk.) Philips purchased both from HP, which has an OEM deal with Hitachi. "We upgraded to two arrays," Windischman says. "And we can plug another array into our tier 1 array if we need more storage."

Windischman says his monolithic system required placing all data on most expensive Fibre Channel disk. Now, he says about 25 percent of his data is on tier 1 with the rest on cheaper disk. "We define tiers by availability," he says. "What we see as mission critical that could never go down is tier 1. We don't want to lose mission-critical performance for those applications. But when it comes to development systems, we can sacrifice some performance and availability."

Windischman says he's still working out the internal pricing for his users. "We know how much the disk costs, but we have to factor in how much it costs to manage it."

Philips also realized savings from reducing the footprint of its arrays in the data center. "We went from six tiles to three, so we cut our footprint in half and we use less power," he says. "There are a lot of ways to justify [the upgrade]."

Dave Raffo, News Editor, Byte and Switch