Calculating Electrical & Mechanical Loads For The Data Center
According to Ron Hughes, president of the California Data Center Design Group, 95% of existing data centers are insufficiently powered to support modern technology.
APC lists traditional data center densities in the 40 to 80 watts per square foot range; however, most new data centers are being built to handle from 150 to 200 watts per square foot.
As a provider of data center design services to some of the most critical and biggest data centers in the United States, Hughes says in addition to more complex power and cooling requirements, modern data centers must also take into account other important factors such as telecommunications cabling and the actual weight of all the equipment.
"When we talk to people about building a new data center, they say their old one was 70 watts and we tell them they are going to need 150 watts. They're shocked. Not only is it twice the density, it's also more expensive," says Hughes.
There are a number of ways to estimate a data center's electrical and mechanical loads: documenting device and environmental controls, using vendor nameplate information, and looking at historical trends for predicting growth.
Although meticulous documentation of a data center's contents is helpful, Hughes says that information helps very little when capacity planning. Likewise, relying on nameplate power stats can also be grossly inaccurate. "What's listed on the nameplates is typically much higher than the actual requirement," says Hughes. Actual loads are often discounted as high as 50% off nameplate data, he says. Relying on nameplate power information can result in an overpowered environment.
The reality is there is little uniformity in data centers. Using a single average of electrical/mechanical loads leaves too much room for error. A rack containing server blades will consume more electrical power thus generating more heat than one housing only patch panels using little or no power.
When designing a data center, California Data Center Design Group develops a "power profile" by defining the power and cooling requirements for individual servers and racks. To increase the accuracy of its estimate, the group creates multiple profiles. For example, a low-tier rack profile may be rated at 3kW, a medium profile at 5kW, and a high profile at 10kW. The load is then calculated based on the racks' profiles.
Planning for future growth is critical when calculating electrical and mechanical loads for the data center. Not enough can end up costing more money in the long run, but the possibility of losing business and overcompensating creates a waste of resources.
Forecasting growth is easier when there is historical data. Examining how power requirements have changed over the years provides a good picture for future needs, that is assuming past performance is an indicator of future growth, which isn't always the case.
Using the existing load of the data center as a baseline, future growth can be calculated based upon the technology that will be implemented in the following 12 to 18 months. "You really need to take a close look at the technology you are moving toward-are you going to be a 1U shop, installing blade servers, or running mainframes?" says Hughes. He also says data center projects are typically long-term-often three to four years-meaning that load requirements are often transitioning from older to newer technology, further complicating power load estimations.
Even more challenging is designing new data centers from the ground up when there is no historical data on which to rely. Again, Hughes' business uses power profiles to estimate electrical and mechanical loads. Advances in power technology, such as the scalable UPS, have become attractive options for new data center deployments. "If you miscalculate, power requirements can be adjusted on the fly," Hughes says.
Although there are methods for making good estimates, there is no set formula for determining electrical and mechanical loads for the data center. From his experience, Hughes says, "It's completely different with every data center."