Designing the Modern Data Center
Businesses are expanding, consolidating, and building modern new data centers to handle new computing demands.

By Darrell Dunn
InformationWeek

Most companies have a data center, even if it's only a closet with a couple of servers. And most companies, including those that started out small, have watched as their data centers grew in size and complexity, consumed more electricity, threw off more heat, and required more cooling as they became vital hubs.

Now a number of businesses are rethinking their data center strategies. Some need to provide more space for servers, while others want to consolidate so they can simplify systems management and increase usage rates with tactics such as grid computing and virtualization.

Business technology professionals who aren't tinkering with their data centers now will need to soon. In a survey of 178 members of AFCOM, an association for data center professionals, more than two-thirds anticipate they'll have to expand their data centers or turn to outsourcing to meet demands in the next decade.

More IT departments need to think about the kind of data center challenges they'll face in the next several years, according to AFCOM's Data Center Institute, a think tank dedicated to improving data center management and operations. It predicts that over the next five years, power failures and limits on power availability will halt data center operations at more than 90% of companies and that by 2010, more than half of all centers will have to relocate or outsource some applications.

That's not news to Jaime Man, director of IT for Health Care Excel, an Indianapolis provider of processing services for health care companies. Man joined HCE about two months ago, and his first project was to create a plan to move the company from an undersized data center to one that can better handle growing demand and accommodate state-of-the art equipment.

The center is in a 500-square-foot area without a raised floor that's populated by about two dozen single-unit rack servers and a dozen old tower servers. It has a single-unit cooling system and an outdated fire-suppression system.

To handle company growth and increasing requirements to store electronic medical records, while at the same time provide quick access to them, HCE needs more servers and storage systems than can fit into the existing facility, Man says. "This current data center is really nothing more than an office they stuck some equipment inside," he says. "We will need to build a true data center environment."

His plan includes blade servers and virtualization. "Instead of having 100 separate servers, I can build an environment with 10 to 15 robust systems, or a few blade enclosures," he says.

Data center consolidation--not expansion--is what Trinity Information Services, a Novi, Mich., provider of IT services to hospitals, needed to grow and secure new business, says Sheryl Zernick, data center operations manager. Trinity began its effort in 1999 by creating a 12,000-square-foot data center at its headquarters. Around 9,000 square feet is being used; Trinity from 1999 through 2004 consolidated separate data centers located in seven Mercy Hospitals in Michigan into the Novi facility. Last year, it consolidated data center operations of four additional Mercy Hospitals in Iowa at Novi.

"When we rolled out the first effort in Michigan, it was a little bump, because it was a new process," Zernick says. "But with each one, it got a little better, smoother, and faster. By the time we got to Iowa, it really moved flawlessly."

Recently, Trinity contracted to handle IT operations in four Holy Cross Medical Center hospitals. Those hospitals are in Fresno, Calif.; Silver Springs, Md.; Columbus, Ohio; and South Bend, Ind., providing a new challenge of integrating operations across the continent. The Novi data center is mainly filled with IBM Unix-based servers that were acquired using "asset swaps" that retired older equipment from the closed centers.

Intel's Consolidation

Then there's Intel. The world's largest producer of chips for PCs and servers is just starting to plan for a multiyear consolidation project for its data center. Intel has around 70,000 servers in about 100 facilities worldwide, including 50,000 servers in about 50 data centers dedicated to meeting the demands of local engineering groups. It wants to combine most of those operations into a more manageable group of around 12 data center hubs, while still keeping enough computing resources locally to help engineers with delay-sensitive applications.

Consolidation is designed to let Intel cut equipment purchases, increase server utilization rates, and develop standardized systems and database management platforms. "The goal is to drive down the number of applications in the environment, consolidate out some old stuff, and drive standardization across different parts of the company," says CIO John Johnson. It also should let Intel boost server usage rates in the engineering environments from about 60% to 80%.

The first part of the process was recently completed in Albuquerque, N.M., where Intel repurposed an old 68,000-square-foot processor manufacturing plant into a data center hub. The plant offered floor space as well as existing power substations and cooling infrastructure.

Intel plans to expand in phases, the first of which created an 8,400-square-foot area for about 1,500 blade servers. The company plans to add 6,400 blade servers soon; a second phase would provide capacity for another 12,000 servers.

"We're looking at this as a blueprint," says Elwood Coslett, director of strategic capabilities in Intel's IT organization. "Today we use a lot of localized and internally written tools, but as we move toward a grid environment, we will need more robust types of services we can operate and manage across the environment." It will take as long as four years to identify and implement standard processes across the hubs as they're built.

Once several large hubs of grid computing are established, Intel will seek to break down the computing demands into small tasks. The majority would be done on the grid, while certain tasks--those most vulnerable to delays--would continue to be completed locally.

Johnson estimates Intel will spend between $5 million and $10 million over the next two years developing the grid computing environment's design.

Other companies are developing similar plans. In the AFCOM survey, 37% say they plan to implement some form of grid computing or virtual processing within two years, 17% within five years, and 4% within 10 years.