BY Jason Togyer - Sat, 2010-05-01 20:05
- Computer science and robotics researchers are searching for practical ways to reduce carbon emissions and energy use
In a four-by-six-mile rectangle of a dusty mountain range about one hour east of Palo Alto, Calif., sits the Altamont Pass Wind Farm, a collection of nearly 5,000 power-generating electric wind turbines.
Since 1981, the Altamont Pass Wind Farm has been a landmark for tourists and--as the largest collection of wind turbines in the world--a symbol of the promise of harnessing the wind for renewable, clean electric power.
It's also very badly designed.
For one thing, many of the turbines are too low to the ground and too closely spaced. For another, air currents in the valley are unreliable, meaning the turbines regularly slow or halt, so the electricity they generate must be supplemented by natural gas power plants. And the turbines kill about 4,700 birds each year, including about 70 golden eagles and 1,300 other raptors, because the Altamont Pass Wind Farm was built smack in the middle of a major migratory path.
In short, Altamont Pass--as one environmentalist told the San Francisco Chronicle--was "the exact wrong place" to build a wind farm. But what's the "exact right place"? And what is the "exact right," or optimized, configuration for the arrangement of the wind turbines?
Jaime Carbonell is no meteorologist or environmental scientist. Much of his research was focused on machine translation and language technologies. But a chance encounter with a utility-company executive during a long airplane flight inspired him to ask questions about wind energy.
It turns out that three decades after the construction of the Altamont Pass Wind Farm, the computer models used to design them remain crude at best, says Carbonell, head of the Language Technologies Institute in the School of Computer Science. He also was surprised to learn that many power companies lack in-house expertise and rely on turbine manufacturers to help them plan wind farms. "Each one does an 'impartial' study and concludes its products are the best," Carbonell says.
Some preliminary investigation left him wondering if the optimization methods he was using on large passages of text could be used to optimize wind farm design. Now, Carbonell is working on just that in collaboration with Jay Apt, distinguished service professor in the Department of Engineering and Public Policy and executive director of Carnegie Mellon's Electricity Industry Center.
It's a new area of research for Carbonell, but one he finds very attractive as a computer scientist. "Very often, the distance between 'theory' and 'practice' is very long," he says. "This is a problem that matters. This is a case where some of the optimization techniques we're using can make a real difference."
He's not the only person in SCS working on what might be called "green" technologies. Heightened awareness of global climate change--and the mounting evidence that human production of greenhouse gases is responsible--has spurred other researchers to see where computer science might help reduce our carbon footprint.
Some SCS faculty members, like Carbonell, are tackling environmental problems using computer models. Others, such as Jen Mankoff of the Human-Computer Interaction Institute and former SCS Dean Jim Morris, now working on the West Coast, are deploying web and mobile phone apps designed to encourage better behavior.
And a few, including Illah Nourbakhsh and Gregg Podnar of the Robotics Institute, are getting their hands dirty--literally--by launching a grassroots effort to convert gasoline-powered cars into sophisticated electric vehicles.
In some cases, green projects are showing benefits besides saving energy and cutting down on greenhouse gas emissions. Take David Andersen's effort to design less electricity-hungry servers using slower processors and low-power storage devices. Rewriting code to run on these so-called "wimpy nodes" not only saves electricity; the same techniques enable that code to run much more quickly on more powerful processors. "It was a very nice dividend," says Andersen, an assistant professor of computer science.
In September, SCS launched a new monthly series of seminars on Sustainability and Computer Science as a forum for researchers to discuss ways in which computational thinking could be applied to energy conservation and the environment. The forums also provide opportunities for School of Computer Science personnel to share ideas with colleagues from other parts of the university, including the School of Architecture; the departments of Engineering and Public Policy and Civil and Environmental Engineering; and the Steinbrenner Institute for Environmental Education and Research.
Behind the seminars and the research activity is a new sense of urgency over climate trends that scientists say are both obvious and alarming. By examining ice samples dating back approximately 650,000 years, the National Oceanic and Atmospheric Administration estimates that current atmospheric levels of carbon dioxide--a byproduct of burning fossil fuels and one of the main greenhouse gases--have never been as high as they are today. Over the past decade, NOAA says, ocean levels have risen at a rate twice that of the previous 100 years; polar ice caps have steadily shrunk in area and thickness; and storms have grown in intensity and volume.
Mankoff, an associate professor in HCII, remembers reading about those and other climate change "tipping points" and coming to a sudden realization "that I could have a bigger impact on these issues by bringing them into my research." With several colleagues, she developed a website called StepGreen.org, which suggests "green actions" that users can incorporate into their daily lives and measures the impact of their energy use. StepGreen also functions as an umbrella system for deploying new ideas, including sensors and mobile devices for helping its subscribers make better decisions about green choices.
"I have two young kids, and a job that takes me away from them a lot," says Mankoff, who helped organize the SCS seminars on Sustainability and Computer Science. "I always want to be able to tell my kids when I leave them that I'm working on something important enough to leave them."
There are flinty-eyed financial reasons for doing sustainability research as well. At a campus-wide workshop Feb. 24, Rick McCullough, CMU's vice president of research, noted that the federal stimulus package alone has allocated $4.4 billion to the U.S. Department of Energy for research into renewable energy. Other state and federal agencies are also funding research into sustainable power generation.
The private sector is also interested. In 2006, the last year for which data is publicly available, servers and data centers in the United States consumed 61 billion kilowatt-hours of electricity in 2006 at a cost of $4.5 billion. Reducing that energy consumption even 10 percent would mean big cash savings. No wonder that cloud-computing giants such as Google and major chipmakers such as Intel and AMD are funding research--at Carnegie Mellon and elsewhere--into ways to reduce their energy consumption. In February, during the first-ever round of Google Focused Research Awards, Andersen and Mor Harchol-Balter, associate professor of computer science, receive $100,000 to study ways to build energy-efficient computing clusters.
"The bottom line is that there are opportunities in energy research, and they're growing," McCullough says. "We have some really fortunate strengths in the environment, solar power, electrical storage, carbon-capture technology, wind and water energy, policy issues--there are excellent, interesting things going on here at Carnegie Mellon."
Wind energy represents one of the biggest research opportunities. A 2006 report by the U.S. Department of Energy suggested that 20 percent of the nation's electric power needs could be met with wind by 2030. But under that scenario, new wind power installations have to come on line at a rate of 16,000-megawatts per year from 2018 through 2030. While engineers have learned from mistakes made at Altamont Pass and elsewhere, finding a location for a wind farm and then placing the turbines in a way to maximize the amount of energy generated isn't a simple task.
Wind farm designers must be mindful of local and state environmental regulations, ground conditions, access to power lines--and of course, the prevailing winds, says Paul Copleman, a spokesman for King of Prussia, Pa., based Iberdrola Renewables, which operates 25 American wind farms, including the Casselman Wind Power Project in Somerset County. The 34.5-megawatt generating site on a former coal mine 60 miles southeast of Carnegie Mellon's Pittsburgh campus was dedicated in 2008 and is one of Pennsylvania's newest wind farms. "You obviously want to maximize the number of turbines you can fit on the land you've leased," Copleman says. "By the same token, you can't just clutter up the landscape with turbines."
Understanding wind conditions, Copleman says, can require a year's worth of monitoring and three or more sensor stations to provide continuous sampling of information that's unavailable from public sources. "It's very valuable data," he says.
But those sensor stations aren't cheap to operate, says Carbonell of the LTI, so while adding more monitors obtains more accurate data, it also increases the cost of the project and can delay construction. Building a computer model that approximates wind conditions using only a handful of monitoring sites would save both time and money, Carbonell says.
Such a computer model has to account for many other variables as well, he says. While the population of the United States tends to be concentrated along its coasts, the strongest, steadiest winds are found in the sparsely populated Dakotas. Wind farms need to be close to trunk power lines, but far enough from major cities that large tracts of property can be acquired affordably, away from nearby residential areas. And they need to be away from flight paths for both migratory birds and aircraft.
All of these variables can be optimized, and although wind farms have little in common with machine translation, designing a computer model employs many of the same techniques, Carbonell says. "In machine translation, we're often trying to align words, phrases and texts in different languages," he says. "You don't align them in isolation, you align them in context."
Wind farm design is a similar multi-variable optimization problem, Carbonell says. "You need to be able to provide the most power at the least cost. If the price of land acquisition goes up, it changes one part of the equation. If the distance to the power grid changes, it changes another part of the equation. So while these are completely different problems, the same tools can be applied."
Tools--physical ones, such as wrenches and screwdrivers--are exactly what Nourbakhsh and Podnar have in mind at the Robotics Institute's Community Robotics, Education and Technology Empowerment Lab. The CREATE Lab is home to a new open-source project to get both "code jockeys" and "gear heads" together to develop a practical process for converting gasoline-powered automobiles into electric vehicles.
Dropping an electric motor and some batteries into a second-hand car isn't a new idea, says Nourbakhsh, associate professor of robotics and co-principal investigator of the Charge Car project with Podnar, an RI program manager. "There are tons of people who do it," Nourbakhsh says. "And there are tons of people who do it--and fail." Many of those projects flounder because adding batteries to increase the range of the vehicle also adds weight while subtracting passenger and cargo space.
Nourbakhsh has first-hand experience with such conversions--since 2001, he and his wife have been driving a Toyota RAV4 that runs on straight electric power. But more important experience might have come while he worked on his bachelor's degree at Stanford, where he helped drive a solar-powered vehicle from Orlando to Detroit using only the sun for energy. "We were optimal controllers," he says. "We paid attention to motor control regulation, trying to predict where the hills were, and we paid attention to where the sun was on the highway."
Optimal control is a goal of Charge Car's prototype, built into a 2001 Scion. It uses an extra high-density capacitor--or "ultracapacitor"--to store electricity and lengthen the life of a vehicle's batteries by handling heavy current draws, such as when the Scion climbs a hill or accelerates away from an intersection. Charge Car is about to launch a monthly contest challenging developers to create intelligent, open-source power-management software that learns a person's driving habits and adjusts the charging and discharging of the ultracapacitor to maximize battery life and vehicle performance.
Using GPS devices, Charge Car is also collecting daily data on real commutes from all over the United States. That's already provided some surprising information--such as the fact that vehicles used on urban commutes spend up to 35 percent of their time decelerating, and not 3 percent, as older models assumed.
That's not the only type of crowdsourcing envisioned by Charge Car. It's reaching out to Pittsburgh-area auto mechanics, vocational teachers and electronics and auto hobbyists, tapping their knowledge to help create an optimum electric vehicle conversion kit that can be manufactured and installed in late-model cars for under $6,000.
"Pittsburgh is perfect for this kind of project," Nourkbaksh says. "We have foundations who are interested in this, we have the university community and we have this incredible industrial background. And Pittsburgh has hills and different kinds of weather, so if Charge Car works here, it will probably work in 80 percent of the country."
On a smaller scale in terms of size--but with potential to make a big impact in computer science--are the so-called "wimpy nodes" being tested by Andersen. Servers in data centers use a tremendous amount of energy, he says, but much of that power is wasted. Speedy processors consume more power to perform operations than slower processors that can perform the same work, and those speedy processors spend a lot of time waiting for memory, or worse yet, for conventional hard drives to access data, especially when they're engaged in jobs that require a lot of searching.
Yet a lower load on a processor doesn't directly equal lower power consumption. Even when a modern CPU is operating at 20 percent of its capacity, it's still drawing more than 50 percent of its peak power.
The architecture Andersen and his team are working on, called FAWN (for "a Fast Array of Wimpy Nodes") uses less-powerful processors that are individually more efficient and reduces the amount of time those processors are idling. Instead of reading and writing data to a hard drive, FAWN caches part of the data in the memory of each "wimpy node," where it can be retrieved in a fraction of the time and with a fraction of the energy that accessing a magnetic platter would require. FAWN further improves its energy efficiency by using solid-state Flash storage, which consumes about a tenth of the power of a hard drive and is up to 175 times faster at retrieving random blocks of data.
Tests have found that FAWN can sort a gigabyte of data six times more efficiently than a conventional rack-mounted server. The biggest FAWN cluster built so far had 25 nodes, and Andersen says the team is working on a 100-node cluster.
Yet getting existing software to run on FAWN isn't always easy. Many programs are tuned for speed, not for conserving memory, Andersen says. "Modern programmers expect a lot of memory to be available, and when you present them with a node that only has 256 megabytes of RAM, a lot of things don't work," he says. Researchers tried to run a popular virus-scanning program on a wimpy node, only to see it run out of memory when it got to its 1-millionth virus signature.
When the code was rewritten to allow it to run on a "wimpy," it also performed two to three times faster on a conventional desktop machine, Andersen says, which was a happy discovery. But FAWN needs to be able to run existing software without requiring major modifications to the code, and that's the next hurdle for Andersen and his colleagues. "We need to be able to provide a black-box solution to current problems, and we're not there yet," he says.
Any green technology that inconveniences developers or end-users will have a hard time finding acceptance, says Morris, currently on a sabbatical while working at Google in Mountain View, Calif. "Like many people, I've gotten deeply interested in climate change, but I've also done some philosophical inquiry," he says. "I've asked myself, 'What are people going to do to change their behavior to limit CO2 emissions?' And my conclusion was 'not much.'"
Part of the problem, Morris says, is that it requires a sense of altruism that most of us can't manage. "The people who are going to benefit from me changing my behavior haven't been born yet," he says.
It's better, perhaps, to avoid the problem by offering people green solutions that offer them a clear, immediate benefit. One of Morris' professional and personal interests for more than 30 years has been encouraging people to carpool to work, but such efforts tend to stall, he says, because carpooling requires a high level of personal commitment. Commuters don't want to be computer-matched with a stranger who's going to share their car forever, nor do they want to be tied to that person's schedule.
Morris has been working on one possible idea he calls "SafeRide." Based on similar technologies to those that allow high school classmates to discover each other on Facebook and get directions from Google Maps, SafeRide would exploit email, cell phones and GPS to match passengers with drivers in real time. A driver might not be willing to share her car with a random stranger five days a week, Morris says. But she might be willing to share her car with someone once in a while if they personally reached out through a networking application.
With a web-capable cell phone, Morris says, SafeRide might even be able to match two people who are both at the airport and need to share a cab to a hotel. "We call it 'high-tech hitchhiking,'" he says.
To Mankoff, those and other important applications demonstrate the different roles that computer scientists can play in reducing greenhouse gas emissions.
"There are all sorts of interesting, hard, technical problems that need to be solved," she says. "The only way the country as a whole is going to address these problems is by getting everyone's help, along with the unique tools and knowledge that each of us brings to the table."
Jason Togyer is managing editor of The Link. He wrote the cover story about the Gates and Hillman Centers for the Winter 2009-10 issue.
For More Information:
Jason Togyer | 412-268-8721 | jt3y [atsymbol] cs.cmu.edu