This article is part of a VB special issue. Read the full series here: Intelligent Sustainability.
One of Microsoft’s largest data centers sits near the Columbia River in Quincy, Washington. Everyone loves the scenery, which is breathtaking, and the rural ambiance, which is a welcome respite from Seattle and Bellevue. The accountants, though, love the fact that the local electrical power is cheap because it comes from hydroelectric dams.
And there’s one more thing: Hydroelectric power is also considered to be one of the greenest forms of energy with a very low carbon footprint. That’s why some call Microsoft’s compound one of the greenest data centers ever.
“This is actually one of my favorite places in the world,” said Brad Smith, the president of Microsoft, when kicking off a video tour during the pandemic lockdown. “Why? I think it represents the most important infrastructure of the 21st century.”
Smith sees the massive data center with close to half a million servers in more than 20 buildings as “an incredible intersection between digital technology, energy technology, environmental science and the need for innovation.”
The company started construction in 2006 and has been expanding the footprint ever since. The lure of low prices from hydroelectric power may make the CFOs happy, but the marketing team enjoys celebrating the low carbon footprint.
What makes a data center green? Building one is a challenging mixture of architecture, network science and heat transfer. Many companies, not just Microsoft, are also asking how they can do a good job on environmental questions too.
It’s not hard to understand why. Some believe that a green sales pitch can attract and keep customers. Some just think it’s the right thing to do. They want to build the best green data centers because they feel that the world will demand it.
Still, a major challenge is trying to understand just what makes a data center green. Some factors, like carbon footprint, are straightforward, even if they’re not always simple to measure. Other factors are more philosophical, and the companies can make elaborate or sometimes strained arguments about how their strategy is good for some part of the environment.
Microsoft’s big data center is one of the easiest to embrace. Hydroelectric power’s low price makes it a popular choice with the CFOs, but it also comes with a nice environmental bonus because no carbon dioxide was emitted into the atmosphere when creating the electricity. The sun evaporates the water and then when it falls into the mountains and rushes down the valleys, some of that energy can be captured by the massive dams that the U.S. built along the rivers in the northwest.
Some, though, are starting to point out issues caused largely by the massive size. Data centers turn electricity into heat and getting rid of the heat isn’t always easy. Some biologists, for example, are protesting that dumping too many kilocalories into the water of the Columbia River is distorting the ecosystem.
Further down the Columbia River near The Dalles, Oregon, Google built a data center for many of the same reasons as Microsoft. Now the company and some residents are arguing about how many gallons a day that Google should be allowed to use to cool the computers.
The dispute has unfolded over a number of years and much of it revolves around non-environmental concerns like whether the city should disclose the details of any agreement to the public.
Google has also found itself in similar disputes in Arizona and Texas. The company is addressing the conflicts both legally and by making public promises. In their Water Stewardship paper, the company sets forth principles for using the water efficiently and responsibly. For instance, some of their data centers recirculate water multiple times, effectively saving water over the past practice of using the water for only one cycle of cooling. Other data centers use air cooling, eliminating water use altogether.
The company has also set a goal of being “water positive” by returning even more water than their offices consume. They explicitly promise to “replenish 120% of the water we consume, on average, across our offices and data centers and help restore and improve the quality of water and health of ecosystems in the communities where we operate.”
Some supporters of the data centers also put the water use in historical context. They point out that cooling modern electronics usually requires much less water than the aluminum smelters that used to be the dominant industry in the Columbia Gorge. Extracting aluminum from bauxite is an energy-intensive process, and companies like Alcoa have long been drawn to Washington State for access to the electricity. Lately, though, Silicon Valley companies are driving out the old metal companies by outbidding them for electrical power.
Deploying artificial intelligence
Water use, of course, is only part of the debate. Google is one of the leaders in trying to control electrical use and put a limit on their carbon footprint, not just in their data centers, but also in their offices around the world. They are deploying advanced algorithms, sometimes using artificial intelligence, and creating partnerships with green energy companies.
“We aim to operate on carbon-free energy 24/7 by 2030 at all our data centers, cloud regions and campuses across the globe – the first company of our size to set this goal,” explained Corina Standiford, a member of Google’s sustainability team. “While it’s important to eliminate our carbon footprint, it’s best if we can use less energy in the first place.”
Google speaks openly about many projects. For example, one of the challenges of using some renewable sources like wind power is the supply varies with the weather. Google turned around and trained a machine learning algorithm to predict the wind, something that’s proven to be accurate over about 36 hours.
The predictions from the AI are then used to schedule electricity purchases from the larger grid when wind power will not be available. Predicting the needs so far in advance gives other power sources the chance to plan, reducing the costs. Now, their cloud data centers can make longer-term commitments to electrical usage and notify non-wind generators in advance if the wind won’t be blowing.
“We can’t eliminate the variability of the wind, but our early results suggest that we can use machine learning to make wind power sufficiently more predictable and valuable,” said Sims Witherspoon, a project lead of the Deep Mind AI tool, and Will Fadrhonc, the lead of the Carbon Free Energy Program, in a blog post.
One of Google’s solutions is to support other green sources with both technology and purchase agreements. One of their investments is in Fervo, a company that makes geothermal electrical generators that are always on. The output will help sustain one of the company’s data centers in Nevada.
The company is also working directly on improving the technology. The engineers have installed a network of fiber-optic cables near the geothermal plant and Google is training their AI to predict how the heat flows in the ground. They hope to use this data to optimize production and also predict the best times when geothermal energy can replace renewable energy sources.
“This collaboration also sets the stage for next-generation geothermal to play a role as a firm and flexible carbon-free energy source that can increasingly replace carbon-emitting fossil fuels,” said Michael Terrell, the Director of Energy at Google in a blog post about the effort.
Another important strategy for Google is to calculate useful metrics about their energy profile and set goals around them. When contacted for the article, Google’s spokeswoman rapidly cited a number of facts that revolved around numerical goals:
- During 2020, all of Google’s data centers delivered “67% round-the-clock carbon free energy”.
- Five data centers are “operating at or near 90% carbon free energy.”
- Google tracks power usage throughout the year and makes sure to include all overhead. This allows them to calculate a system wide “power usage effectiveness” of 1.10.
Their goal is to turn these numbers into marketing that can help sell users on switching to Google’s Cloud Platform. When companies are looking for cloud instances, they’ll include the numbers in the choice matrix like the Power Usage Effectiveness (PUE) or the amount of Carbon Free Energy (CFE).
The Google Cloud Region Picker is one tool for letting customers choose to move their workloads to the greenest Google data centers. Overall price and latency, two values normally used to help select cloud providers, are placed right next to the carbon footprint. At the moment when I was writing this, the Cloud Picker steered me to their servers in Iowa where the price was $0.021811 per virtual CPU hour while the carbon-free energy score was 93%. The site also calculated that carbon output was 454 grams of CO2 equivalent per kilowatt-hour.
Google’s marketing efforts show that their customers are willing to consider the environmental effects most often for non-interactive work that would run in the background.
“For best-effort workloads like batch jobs or backup, carbon scores were ranked as the top characteristic more than any other factor,” said Chris Talbott, from Cloud Sustainability, and Steren Giannini, a senior product manager, in a blog post.
Bitcoin to the rescue?
Sometimes the question of carbon footprints and green computing is a bit more complicated. One of the most difficult philosophical knots to unravel is the environmental value of some local Bitcoin mining operations that are parked next to oil wells. Lately, both environmentalists and technologists have complained that Bitcoin mining is incredibly energy intensive, and some estimates suggest that the blockchain miners consume more electricity than a small European nation. The miners do no useful work beyond running an elaborate mathematical game to establish consensus.
But what if the energy powering the mining had been wasted – or worse, vented into the atmosphere where it could sit for decades absorbing heat as a greenhouse gas? A number of companies are starting to build mobile data centers by filling shipping containers with computers. Then they drive them to local sources of low-cost energy. One example, Giga, brings specialized hardware for mining bitcoin and parks it near oil wells where it runs the mining computations using energy harvested from burning the natural gas in a generator instead of burning it in an open flare.
The developers of these data centers argue that their Bitcoin mining solutions are greener, at least, than the alternative. Many drilling operations end up releasing natural gas as a by-product and they often burn the gas in open flares because there’ often no easy or economically feasible way to capture it and bring it to market.
The mobile data centers divert the flare to a local generator that burns the gas to produce electricity. Instead of being wasted, it’s put to use. Not only that, but the engines powering the generators usually do a better job of burning the gas than the naked flares. Little methane escapes into the atmosphere. This may not be as green as, say, using renewable energy, but the operators of the small, portable data centers still claim to be improving the health of the planet.
Bitcoin mining is an easy option for these data centers because the algorithms don’t need much interaction with the larger network. The shipping containers can be moved to remote locations with very low bandwidth connections from cellular networks or satellites.
In the future, these mobile data centers could also handle other compute-intensive work that doesn’t require fat or fast connections to the larger Internet. Some scientific tasks like simulating protein folding, for instance, might also be ideal.
Far-flung data centers
Not all of the distant data centers rely upon natural gas for their electricity. One operation in Kenya relies upon geothermal energy extracted from some of the geological fault lines in the Great Rift Valley. The opportunities are great, and some estimate that there may be many thousands of megawatts of untapped power.
In one report, the state power company, KenGen, is said to be reaching out to Bitcoin miners to bring their operations to the Rift Valley. They believe that Bitcoin mining is easily moved to Africa so that the untapped geothermal energy can be converted into digital currency. Geothermal power is believed to be very environmentally friendly because it does not release carbon dioxide into the atmosphere.
Large versus small
An interesting debate is emerging about whether small or large data centers are greener. On one side are the big cloud companies that tout the ability of the large data centers to embrace the latest technologies and exploit all of the economies of scale. Microsoft’s big data center on the Columbia Gorge, for instance, runs on hydroelectricity, but when it needs other power, it uses diesel generators.
Microsoft wants to make even this usage greener. They’ve committed to using the best diesel generators that run on biodiesel and only release the well-filtered exhaust. They also intend to phase out diesel by 2030.
Many smaller data centers, especially the ones that are built by companies in their office buildings, are often afterthoughts, and the local companies often cannot use the best practices. They may not have very efficient air conditioning, nor can they use the greenest electric options. The fans of the hypersized data centers point to their age and inefficiencies.
But sometimes the small data centers have other options. In the wintertime, a local data center can use the extra heat to keep the rest of the office space warm. The extra square footage devoted to the data center may simply be surplus space that the corporation was already paying to support. Simply moving the data center from the basement of the local office may not save as much as it might appear.
Fear and financial worries
For all the easy excitement, though, it’s important to keep in mind that many data centers do not have green goals or saving the environment at the top of their list. Where the solutions are easy and come with often substantial cost savings at the same time, as they do with adopting hydroelectric power, the companies are quick to embrace green power.
But when the cost of alternative energy drives up the price of electricity, many aren’t so openly enthusiastic. Cloud costs remain a major part of IT budgets and, in many cases, they’re already growing because companies are embracing the clouds for other reasons. Adding in still higher prices for greener options is often met with resistance from CFOs in budget-minded firms.
This is obvious after a bit of research. While all of the cloud companies are happy to speak about their performance and pricing, many don’t have much to say about any green initiatives. They feel that price is still the most important factor for their customers.
Several cloud companies, both big and small, refused to discuss the matter on the record. One said simply through their public relations firm that they wanted to “pass on this for now, as they don’t have much to contribute.”