Company Keeps Its Cool, Lowers Costs

By CIOinsight  |  Posted 04-27-2007

Company Keeps Its Cool, Lowers Costs

The research firm Gartner last fall ranked Data Return LLC, a small, independent managed services provider, among a handful of leaders in its 2006 North American Web Hosting Magic Quadrant report.

Gartner positioned Data Return ahead of some heavy hitters, including CSC, Verizon Business and Quest Communications International, which also manage information-technology services for customers over the Internet. "It has a track record as a risk-taking visionary with significant engineering expertise and a highly automated infrastructure, including a very good customer portal," the report noted.

Though it's privately held and does not make its finances public, the company says it's had 19 consecutive quarter of profitability, has a client churn rate of less than 2 percent, and has added a number of new customers in recent months, including Burton Snowboards, shipping company Agistix and direct marketing company AB&C Group.

Data Return's current success has a lot to do with how the company responded to several key challenges it was confronting several years ago. Specifically, the company had to find a more effective way of providing processing capabilities for customers such as H&R Block; the tax preparer's business is often seasonal, with the result that much of the infrastructure used to support H&R Block's Web site went unused much of the year. Second, Data Return was seeking ways to improve customer service via a highly available, efficient and high-performing IT environment that would also allow the company to readily roll out new services to its customers.

"For our customers, stability and availability are essential," says Jason Lochhead, the company's co-founder and principal architect. "H&R Block can't be down during tax season, and it doesn't do any good for our services to be up and available if the system is so slow that nobody wants to use it."

Finally, if it was to succeed in a highly competitive market, Data Return had to ensure its services remained affordable, which ultimately meant minimizing power consumption and cooling costs in its data centers.

Next page: First Steps

First Steps


In 2005, Data Return actively began looking for a solution to all these challenges. At the time, it was using AMD Opteron technology for dedicated servers and larger SQL servers. "We started using Opteron as a platform of choice," Lochhead says. "There were a number of reasons for that. AMD was the only one that supported x64 [64-bit microprocessor architecture], and the reasons we chose it for quad socktype configurations were its HyperTransport technology [a high-speed, low-latency, point-to-point link] and the integrated memory control. We received better scalability as we added more processors, so it made sense to use AMD Opteron processors for the larger servers."

With its existing IT infrastructure, Data Return was able to meet customers' needs—even those of seasonal companies like H&R Block—but not as efficiently, or effectively, as Lochhead would have liked. In order to meet those needs, Data Return would have servers in a standby or sleeper mode, and would power those up and let customers use them for whatever time they needed. "That worked, but it wasn't really a very elegant solution," Lochhead points out. "It was also pretty capital-intensive on the customer's part, so we wanted to see if there were some better solutions out there."

With the AMD processors already integrated into Data Return's server farms—the company has computer operations in Dallas, San Francisco and London—Data Return began looking at virtualization as a way to provide a better, utility computing-type offering, according to Lochhead. The virtualization solution was to be based on dual-core Opteron processors running VMware virtual infrastructure software on Hewlett-Packard ProLiant DL385 and DL585 servers. "We'd been partnering with HP for years, and we'd already started with AMD," Lochhead says. "We talked to [AMD] about what the road map was, and we'd already begun working with VMware virtual infrastructure software. I had already been using it for a year in our lab."

In mid-2005, Data Return started the project in one data center, doing a lot of testing and planning up front. "We began with a fairly small implementation but with an eye to expanding it rapidly," Lochhead says. Today, Data Return has adapted virtualization in its Dallas and San Francisco data centers, but has not yet implemented it in the London center. "Technically, it hasn't been that hard to roll out," he says. "The toughest part was a change of culture and a change in the way of doing things for the technical people. We have this virtual layer that enables them to do a lot of things they couldn't do in the past. They have to get used to dealing in that environment."

As a pioneer in virtualized hosting, Data Return was initially worried that the market might not be ready or willing to operate in a virtual environment as well. "We were afraid people would not be willing to run production-level servers in a virtual environment," Lochhead says. "What we found, however, is that a vast majority of customers already have some kind of VMware in their environment, so it hasn't been as difficult to sell as we expected."

Even midsize customers, which normally require less computing power and have far smaller IT budgets than Fortune 1000 companies, are primed for virtualization, Lochhead believes. In fact, Data Return is rolling out a new utility-enabled managed services platform called Infinistructure to pursue the midsize market.

Next page: A Powerful Benefit

A Powerful Benefit


For Data Return, one of the major, and largely unexpected, benefits of virtualization has been a reduction in cooling costs while optimizing server density needs. "When we started looking at energy efficiencies and power costs, they weren't really high on my list of things that were important," Lochhead admits. "But as we rolled it out, I started to realize how big a lever it could be."

Lochhead says that the combination of virtualization and the AMD Opteron processors' low power requirements keep power and cooling costs down, and enable the company to achieve a smaller footprint with greater rack density. Consequently, it can keep prices down—a key factor for survival in the managed-service provider marketplace.

Within the company, data center staffers have become focused on cooling and related cost issues. "The guys running the data center know exactly what's going on, so they understand we really have to maximize those finite resources, especially in the hosting business," Lochhead says. Soon after the implementation, data center staffers started gauging "how many dollars they were getting for however much cooling and power they were expending on a given server," he says.

"Virtualization fits great for that because you're maximizing your space, and you're using the CPU capacity you have in one box rather relying on 10 boxes where you're using only 10% of their CPUs," Lochhead says. "One of the guys in charge of our data center has a big spreadsheet. He did the math and concluded that depending on the type of server we're retiring, what kind of workload it has, and how many servers we can consolidate, we could save between 50% and 70% on our power and cooling bills if we were to virtualize all of our server farms."

The upshot of all this, according to Lochhead: "What we ended up with was addressing more than just the computing requirements for customers. We ended up creating a new platform for ourselves and continue to discover new ways to use virtualization in the data center. It's really becoming our primary platform."