ADVERTISEMENT
  About the SA Blog Network













Plugged In

Plugged In


More than wires - exploring the connections between energy, environment, and our lives
Plugged In HomeAboutContact

Data centers of the future might be their own power plants

The views expressed are those of the author and are not necessarily those of Scientific American.


Email   PrintPrint



The data center of the future might do more than crunch and store information. In addition to serving web pages, streaming Netflix videos, and hosting social networks, data centers might produce their own power.

Rows of servers in an Icelandic data center. Credit: Intel Free Press

Data centers consume a lot of energy. In the United States, data centers used 76 GWh of electricity in 2010 (or roughly 2 percent of the country’s energy usage, according to a report in the New York Times. Therefore, companies are looking for ways to reduce energy usage and pollution while maintaining reliability.

A team of researchers at Microsoft is studying how they could bring the power plant into the data center itself to improve efficiency and reliability using fuel cell technology.

“We are taking an unconventional approach to power a datacenter entirely by fuel cells integrated directly into the server racks. This brings the power plant inside the datacenter, effectively eliminating energy loss that otherwise occurs in the energy supply chain and doubling the efficiency of traditional data centers”, writes Sean James, Senior Research Program Manager for Microsoft’s Global Foundation Services (the group that manages the company’s data centers).

Data centers typically source electricity from utilities (or in some cases provided by onsite renewable generation). If electricity is purchased from the grid, it is likely produced using some form of combustion. First, fuel is combusted to spin a turbine (either directly in a gas cycle or using steam). The turbine blade then spins a generator to produce electricity, which is then transmitted and distributed along wires until it gets to the data center.

The laws of thermodynamics tell us to expect losses at each of the stages, so much so that only 20 percent of the energy produced at the power plant makes it to the server rack. Losses occur every time energy changes form (say, going from chemical energy in the fuel to thermal energy in steam, to mechanical energy of a spinning turbine blade, and so forth).

By eliminating the number of energy conversions, James and his team have found they can double the efficiency to 40 percent.

Fuel cells work by splitting protons and electrons from a fuel molecule (like hydrogen). A membrane allows only protons to pass through. Electrons, looking for the easiest route to travel, are directed through an external circuit, producing electricity.

James and his team have identified two fuel cell technologies, solid oxide and molten carbonate, as candidates for powering data centers. Both can reform hydrocarbons like methane (either from natural gas or biogas) without the need for expensive equipment to condition the gas for use.

A prototype fuel cell and server rack. Credit: Microsoft

The placement and capacity of the fuel cells will depend on the needs of data center operators, but James sees a server rack coupled with a fuel cell (with several kilowatts of capacity). This design eliminates the power distribution system in the data center and replaces it with a fuel network, which also isolates any fuel cell failures to several server racks, instead of an entire server farm. However, a challenge for the team is finding the balance between reliability, cost, and efficiency.

“It’s the classic Goldilocks issue; not too hot, not too cold or not too many servers to pose a greater risk during a fuel cell failure, but not too small resulting in high cost small fuel cells. It is conceivable that the fuel cell could be integrated into the server for even higher application availability and higher efficiency or a larger system powering a group of racks”, James tells Plugged In via email.

The primary challenge will be the ability of a fuel cell to follow a data center’s load profile. Demand can spike in milliseconds while fuel cells are slow to react, taking several seconds to adjust. Just like how a grid operator must balance electricity generation with demand at any given time, a data center will need to manage generation based on server load.

“Fuel cells must have time to adjust gas volume and internal core heat after both positive and negative load steps. While data centers have a fairly consistent load profile, relative to other large load facilities, load changes are much more dynamic at the server rack level as applications respond to user demand. However, the greatest magnitude transients are expected to come from server power cycling.”

A full-scale data center powered by fuel cells is still several years out. Yet, as more services and information are hosted in data centers around the world, designing clean and reliable ways to power them will remain an important focus.

David Wogan About the Author: An engineer and policy researcher who writes about energy, technology, and policy - and everything in between. Based in Austin, Texas. Comments? david.m.wogan@gmail.com Follow on Twitter @davidwogan.

The views expressed are those of the author and are not necessarily those of Scientific American.





Rights & Permissions

Add Comment

Comments are closed.

More from Scientific American

Scientific American Dinosaurs

Get Total Access to our Digital Anthology

1,200 Articles

Order Now - Just $39! >

X

Email this Article

X