Flexible DC: Is IT ready for factory built data centers?

Eighty years ago Sears hit pay dirt when it offered kit homes to consumers. Built in modules at the factory and delivered by rail, the Sears Catalog homes created a standardized infrastructure ready for high tech innovations such as central heating, indoor plumbing and a system for distributing electric power to every room in the house.

In announcing the HP Flexible Data Center this week, HP is applying many of those same concepts to the construction of data centers, which today tend to be highly customized, one-off designs that are expensive to design, build and operate. The result, HP claims, will be data center buildings that can be up and running in half the time, that can be expanded in modules, that run extremely efficiently -- and that carry a price tag that's half what you'd pay for a traditional data center.

Think of HP's concept as moving the industry from the container-based data center in a box to a soup-to-nuts data center delivered in an interconnected series of modular, metal buildings complete with power, cooling and office space. The use of standardized, modular components and designs, and the manufacturing economies of scale achieved by building a data center on an assembly line, could cut new data center costs by 55% to 60%, says Peter Gross, vice president and general manager of HP Critical Facilities Services.

But to get those savings, IT -- and the organizations they support -- will have to give something up. Businesses must be willing to work within the confines of the designs HP offers, and choose from a "short menu" of standard building block options.

Following a trend

HP is, however, following a trend -- and the time may be ripe. Already, virtualization has enforced a standard for deploying new servers, and users have learned to accept that. Demand a physical server or a highly customized virtual server and you'll pay more -- and get it later.

With the move to private cloud, benefits accrue from offering a suite of standardized services on top of that virtualized infrastructure, with an automation layer that provisions, meters and bills back for those services. Again, the economies derive from users accepting the fact that they must choose from a limited menu of configuration options or pay extra for the luxury of customization.

What HP is doing with data center designs builds on that momentum. The idea has merit, but the business is used to having IT deliver everything its way. IT will have to sell the value to the business, just as it must do as it moves to the internal cloud model. And while the data center will surely work well if the equipment within it is all from HP, the designs may not work as well if you're trying to shove a bunch of legacy equipment into that space.

The value proposition

For many companies a new data center is the biggest capital outlay the business will make. Some design engineers argue that the expense is distorting how data centers are built and hosted by favoring scale over more flexible, build-as-you-go designs.

Some of the world's largest data centers cost more than $1 billion to build, so a 50% savings would be huge - if the product catches on and delivers as advertised.

Today a typical 1 megawatt (MW) data center costs about $25 million to design and build. Larger data centers benefit from economies of scale but are still costly: A 10 MW facility typically falls into the range of $17 to $20 million per MW. Gross says Flexible DC will be in the range of $8 million per MW.

Larger data centers have been more efficient to operate because IT organizations have the scale to build in more state-of-the-art, energy saving technologies. That's one reason why many organizations have been consolidating data centers into a fewer, larger ones.

KC Mares, president of Megawatt Consulting, says a bigger is better approach to gain economies of scale in data center construction isn't always the right one. He prefers to build data centers in autonomous "chunks" that can be updated as equipment is refreshed every 3-5 years. "We can't predict what the demand will be or what the technology will look like even a few years from now," he says. So why invest in extra capacity that may be obsolete before it's ever used?

Flexible DC attempts to address that. It gains economies of scale through the assembly line - economies driven through less expensive factory labor. As with other types of industrial buildings, the complexity of data center building designs often run into issues in the construction phase that result in change orders in the field - and the customer always pays for those errors. That leads to costly cost overruns that can add as much as 15% to the total cost of a project. With a factory engineered building most of those issues go away.

While vendors have talked about standardizing aspects of the data center, such as using smart racks that integrate power and cooling, HP is the first to extend the concept to include the four walls enclosing it.

Prior to this vendors also manufactured containers, a data center in a cargo box that functioned as a self-contained appliance - just add chilled water and a critical power source. Flexible DC applies the concept to an entire prefabricated facility. "It's a totally different way of building data centers, says HP's Gross.

Flexible DC basics

the Flexible DC "butterfly" design includes a core unit, the main structure, which includes administrative office, networking and staging areas. Around that customers can add four quadrants of 6,000 square feet each that deliver up to 800 kW for a total of 3.2 MW of conditioned power. Data centers get bigger by scaling out - adding adjacent buildings, each with a core and four quadrants holding IT equipment.

Flexible DC

Data center industrial park: Flexible DC's design includes up to four 6,000 square foot "quadrants" surrounding a core that contains offices. Each five-module structure supplies up to 3.2 MW of power. For a larger data center you'll need room to scale out, adding new Flexible DC structures in cookie-cutter-like fashion. A 20 MW data center would require the six structures you see here. 

The design avoids chilled water in favor of evaporative cooling and air-side economizers. There is no raised floor - all air flows run through a ceiling plenum. A full design can be drawn up, manufactured and installed in about a year, about half the time for a standard data center, Gross says.

HP's approach allows designs to be rolled out in variable sizes. Gross says HP will use hundreds of different suppliers for components and will use its supply chain experience to drive down costs for constructing data centers in exactly the same way it has in building laptop computers.

The long view

Gross says HP's ultimate goal is to cut costs by 75% over what IT pays today to build a data center. "As we build volume we'll bring the cost down to $5 million per MW or less, compared to $20 million today," he says.

Will IT buy in? The momentum is going HP's way, the potential savings look irresistible on paper, but HP is also ahead of the acceptance curve. Sound bytes notwithstanding, businesses still expect to have everything customized to their needs - and they always want it now. But if IT goes with an approach like Flexible DC and starts adding customizations the value will be lost.

IT has always done highly customized applications for the business - it's expected. Ultimately the business needs to decide that, unless there's a compelling, direct competitive advantage from customization, faster and cheaper is always better. That type of thinking needs to be applied from the data center infrastructure level all the way through to the application layer services that IT delivers. And it needs to be enforced from the CEO on down the line.

Some businesses are ready for that. The rest, not so much.

Copyright © 2010 IDG Communications, Inc.

7 inconvenient truths about the hybrid work trend
Shop Tech Products at Amazon