Fueling the Internet
Is the issue quantity or quality of energy?
This is part of IEEE Spectrum's special report: Always On: Living in a Networked World.
Amy Arnold of Menlo Park, Calif., spends much less time at her computer since she left Hewlett-Packard to stay home with her kids. Mostly she uses it for the half-dozen e-mail messages she gets a day. She used to turn it on every few days and spend an hour or so dealing with the accumulated e-mail. But late last year she signed up for digital subscriber line (DSL) service.
Now, Arnold turns her computer on when she wakes up and leaves it on until ready to go to sleep. Even though her computer is in so-called "sleep" mode for much of the day, Arnold's computer usage will raise her household's electric consumption by a percentage point or two. Add to that the increased power usage of all the other consumers signing up for DSL, cable modems, and other broadband Internet connections. Add corporate computer users, who are demanding more powerful machines, more kinds of, and more, peripherals (scanners, for example, once relegated to art departments, are becoming ubiquitous), and faster Internet connections. Then add in the air-conditioned server farms that distribute e-mail, host Web sites, and manage all these connections.
The total impact? It's not clear. Some of the new, more powerful computers make more efficient use of energy than the obsolete models they replace, and the other numbers can vary, depending on how the calculation is done.
Among those who believe Internet use is driving power consumption much higher are Peter Huber and Mark Mills, co-authors of the Digital Power Report newsletter. They calculate that Internet use currently accounts for 8 percent of U.S. electric power demand, and they project it to rise to 30-50 percent in 20 years. Meanwhile, they expect overall demand for electric power to grow 33 percent over the next 20 years.
Huber and Mills, in their analysis, consider the consumption of power by PCs, by servers, and in computer manufacturing. For the past year or so, a lot of attention has been paid to their widely published conclusion that the digital economy is driving demand for power far beyond any projections made in past years by the power industry. They believe it will require the construction of a host of new power plants in the coming decades.
Living in Silicon Valley, this writer last summer became familiar with new terms such as "rolling outage block" and "stage 2 emergency." The first is a system that shuts off power to consumers for short periods on a scheduled basis, and the "stage 2 emergency" means that more than 95 percent of the state's available electricity supplies are in use, and reserves are dangerously low. Local papers urged readers to turn off idle computers, printers, copiers, and other electronics and, whenever possible, use laptop computers and ink-jet printers, which consume less energy than standard computers and laser printers. So the Huber and Mills analysis seemed to ring true.
But is it? The bulk of the experts who have looked at it say no, that the California power problems have a host of causes, but Internet use is not one of them.
A vast overestimate
Jonathan Koomey, a staff scientist for Lawrence Berkeley National Laboratory, Berkeley, Calif., focuses on analyzing the consumption of electricity by computers and consumer electronic products. He told IEEE Spectrum that he believes the Huber-Mills estimate of Internet power use is off by a factor of eight or more, meaning that Internet use currently amounts to little more than 1 percent of U.S. electric power demand. "In our independent technical peer review of Mills's analysis, we compared measured data to Mills's assumptions and concluded that he had vastly overestimated electricity use in every category of office equipment," Koomey said.
Joseph Romm, executive director of the Center for Energy and Climate Solutions, a division of the Global Environment and Technology Foundation, Annandale, Va., takes issue with the Huber and Mills conclusions. He points out that in the four years prior to the big Internet boom (1992-96), growth in the U.S. gross domestic product (GDP) averaged 3.2 percent a year, while electricity demand grew 3 percent a year [see figure]. In the Internet era (1996-2000), he said, GDP growth has averaged over 4 percent a year, while energy demand is growing only 2.5 percent a year. "In addition," he told Spectrum, "the Internet era has also been [climatically] the hottest period in U.S. history. So if anybody tells you that the Internet is an electricity hog," he said, "the data refute that."
Romm, a former U.S. assistant secretary of energy for energy efficiency and renewable energy, thinks there are several reasons why the GDP is increasing more quickly than energy demand. He lists gains in productivity, the replacement of old, energy-hogging computers with more efficient models, and the increasing ability of people to work at home, which reduces both transportation and office energy costs.
"I believe the single most important thing the Internet does is allow better use of existing resources, and the single most unutilized resource is people's homes--they are already constructed, and use a lot of energy whether you are there or not. One hundred square feet of office space uses 2 MWh per year; give me a laptop, let me work at home, and I save a huge amount of electricity."
Chris Lotspeich, a senior associate at the Rocky Mountain Institute, in Snowmass, Colo., said his organization agrees with Romm's analysis. "Structural changes in the economy, facilitated by the Internet, are reducing overall energy use," he told Spectrum.
For Lotspeich, opportunities for energy savings abound. The devices could be made more efficient [see "Plugging the Leak"], components and systems be produced more efficiently, and the buildings that house them be made more efficient at cooling them.
"The biggest problem with computers [referring to the server farms that are the Internet's workhorses] right now is the air conditioning needed for them to operate," agreed Joe Costello, an energy consultant with the Center for Energy Efficiency and Renewable Technologies, Sacramento, Calif. "But there can be better ways to design buildings. And in the long run the Internet is going to greatly reduce our need for energy."
A question of quality
The long-term question may not be one of energy supply, but of energy quality.
"The issue is what I call digital-quality electricity," said Karl Stahlkopf, vice president, power delivery, for the Electric Power Research Institute (EPRI), Palo Alto, Calif. "The silicon revolution has brought increases in productivity, but at a price."
That price is an increased sensitivity to power outages. Transmission and distribution systems for electricity were designed for a world that could weather outages of a minute or so without major inconvenience.
"Today," Stahlkopf said, "a one-cycle, one-sixtieth-of-a-second outage can have very large consequences."
At Sun Microsystems Inc., for example, also in Palo Alto, outages are estimated to cost the company US $1 million per minute, because of the production lost when manufacturing lines are interrupted.
EPRI estimates that nowadays lost productivity due to power outages costs the United States $50 billion annually. And that number, Stahlkopf said, is rising as more and more industrial processes come under computer control.
"At a steel rolling mill, for example," he said, "a power outage knocks the computer system off-line [and interrupts production]. Then everything that had been going through the plant at that time has to be recycled. An outage at a paper manufacturing plant takes one or two shifts to clean up and costs about $250 000."
Reducing power outages is not just a matter of generating more electricity, which could increase dependence on non-renewable fuels [see "An Unnatural Rush to Natural Gas?"], and "EPA's Perciasepe Weighs Impact of Air Regulation on the Energy Mix"]. It may mean that more local generation of power is needed, to reduce dependence on the grid. Additionally, it involves making the transmission and distribution grids more reliable. But, according to Stahlkopf, little investment is being made in these areas.