Information Technology (IT) looks clean. Individually our laptops and mobile phones consume only small amounts of energy and there are no apparent emissions. However, IT is not without environmental impacts. Our high-tech devices require vast amounts of water and energy to create, contain materials that are hazardous to our health and the environment, and are costly to recycle. One example (from PE International GmbH) showed the carbon footprint of a small laptop to be approximately 400kg of CO2, including manufacture and four years of usage.
Apart from the individual devices, however, there is also the overhead of providing the always-on, always-available access to data that we have come to consider the norm. This information is stored in data centers and accessed via the telephone network.
Individual requests for data appear insignificant. For example, Google stated in January 2009 that one Google search “is equivalent to about 0.2 grams of CO2.” It is, of course, the sum of all the requests that shifts our perspective. Imagine for a moment that you could see the information flowing through the wires and the airwaves. Most of us would be surrounded most of the time by vast streams of data, mostly emanating from data centers.
It’s hard to know how much energy is used by data centers, and security and competitive factors mean that owners are tight-lipped. But the United States Environmental Protection Agency estimated that U.S. data centers and servers accounted for 1.5 percent of the total U.S. electricity use in 2006.
Perhaps because of the perception that doing things online is “greener” than doing things in the real world, there has been, until recently, relatively little consideration of the impact of all this computing.
When I interviewed Tom Raftery in June 2006 about social media, he mentioned almost in passing that he was building a hyper- energy efficient data center in Cork, Ireland; it sounded like a rather quixotic undertaking. Tom and his partners at the Cork Information Exchange (CIX) focused on ways to minimize energy use. For example, they used cold aisle containment—completely sealing the cold aisle in the data center so that cold air goes only to the servers. They also used extremely efficient, uninterruptible power supplies (UPS), sourcing units that had an efficiency of up to 98.4 percent (compared with a more traditional 75 percent). CIX now promotes itself as “a leading provider of green, mission-critical colocation services.”
More recently, Sam Ng at Optimal Usability in New Zealand investigated hosting options for Optimal Workshop (their usability software-as-a-service application), and tried to calculate their carbon footprint. Provider RimuHosting persuaded them that using semi-dedicated servers in preference to fully dedicated would be a more sustainable solution. A fully dedicated server would mean having one computer permanently assigned to their application, whereas semi-dedicated would mean sharing the computer with others, with concomitant savings in energy use (and price). According to RimuHosting, a semi-dedicated server would save approximately 1 tonne of CO2 emissions per year, or the equivalent of 24 trees.
Liz Quilty at RimuHosting says the company believes it has “a responsibility to the environment as well as our customers,” although she adds that most clients still don’t inquire about environmental impacts when purchasing.
Of course, large companies like Google, Microsoft, Amazon, and Yahoo, with huge server farms, have a very strong interest in reducing energy costs. This is an area where the dual goals of sustainability and profitability are strongly aligned, so it’s likely that good progress can be made.
For example, Google touts a five-step plan to move toward a more sustainable model:
- Minimize electricity used by servers
- Reduce the energy used by the data center facilities themselves
- Use recycled water instead of fresh water
- Reuse or recycle all electronic equipment
- Engage with peers to advance smarter energy practices
Microsoft has a similar set of steps:
- Use recycled resources whenever practical (including water)
- Use renewable resources whenever available. For example, the Microsoft data center in Dublin, Ireland, uses outside air for cooling (perhaps no great surprise to those of us familiar with Dublin’s climate!)
- Reduce waste in operations (for example, ordering servers literally by the truckload to minimize packaging required when servers are delivered individually or in racks).
- Take part in industry environmental groups.
Many of the steps being taken by data center operators are hardware-related (for example, improving power conditioning to minimize losses and using more efficient cooling).
Microsoft has also identified ten business practices for environmentally sustainable data centers. These highlight the strong connection between environmental and economic goals. The first item on the list of practices is to “provide incentives that support your primary goals.” Microsoft points out that while technologies and practices that support energy efficiency already exist, industry adoption has been relatively low because “data center managers are typically compensated based on uptime and not efficiency.”
There are now many bright people working on the problems of energy efficiency in data centers. James Hamilton, vice president and engineer at Amazon Web Services, in his keynote address at the Usenix technical conference in June 2009, presented a fascinating and practical analysis on power usage in high-scale data centers. He pointed out that as computing power gets cheaper, the cost of power and infrastructure are likely to become dominant. This provides additional impetus to reduce these costs.
Because up-time is such a key measure of success, there has been a tendency to over-specify. Hamilton and others have pointed out that most of the components in server farms are specified to withstand temperatures much higher than the ones to which they are exposed. Reducing air conditioning may often be possible without detrimental side effects.
Utilization is also an important factor. Feeding electricity to servers that are not operating at near capacity is inherently wasteful. Steps such as Optimal Usability’s use of a semi-dedicated server can help reduce waste, but the real solution is for data center design to be leaner, with operators making better decisions about how much computing power is actually needed to meet required service levels.
Besides all of these supply side issues, there are also issues of demand. As designers, we frequently have some control over data calls. For example, I’ve often castigated banks and similar organizations for failing to provide adequate historical data (which have frequently been bound by legacy decisions). On the one hand, I’m pleased to see that in recent years they have made more historical account information available. If we want to minimize server usage, however, we would request a sensible subset, and retrieve additional information only on demand (rather than retrieving and caching for possible display). Thinking about simple problems like this may help us to consider end-to-end issues—not just in physical or visible manifestations, but also in the ways that we store and retrieve data.
Creating efficient data centers with significantly lower environmental impacts is attainable and highly desirable for economic reasons. Hopefully we can continue to be ravenous infovores while treading ever more lightly on the planet