We first examine the ethics of data centers under a utilitarian framework; that is, we look at the use of data centers in the context of their consequences. A couple of notes on this framework: we use well-being as the calculus for weighing the consequences of data centers, and more specifically the well-being of society rather than that of specific individuals. Further, as we consider the ethics of the actions of specific companies building and utilizing these data centers, we look to foreseeable consequences rather than the actual consequences (in the cases outlined below there does not seem to be much distinction between the two.) Finally, we are concerned with act utilitarianism rather than rule utilitarianism. In other words, we look to the specific consequences of the actions firms take surrounding data centers, rather than the general consequences of such actions.
Undeniably, data centers are one of the main drivers of the modern economy. Large data banks are crucial for any business operating in the cloud or with cloud services. These businesses provide valuable services to people globally. The net impact of data centers in and of themselves is almost certainly positive in terms of their consequences to human well-being. Therefore, we focus on analyzing the specific practices companies, especially large companies that provide cloud services to billions such as Microsoft, Google and Amazon, under the utilitarian framework outlined above.
To begin, these data centers require a massive amount of energy usage. One report found that data centers accounted for 1.8% of total U.S. electricity consumption in 2014. Unfortunately, these data centers are often powered by nonrenewable energy sources. In the face of complaints from organizations such as Greenpeace, companies like Google and Facebook have pledged to move to powering their data centers entirely with renewable energy sources. Further, these companies can often act as the impetus for meaningful policy around renewable energy. For example, Google facilitating the creation of a new renewable energy program at Duke, and Facebook’s push for renewable energy in their data centers led to a change from a nuclear power plant initiative to wind energy in Iowa. However, the overall efficacy of this signaling in creating sustainability is questionable. Moreover, a study from the Multidisciplinary Digital Publishing Institute found that this signaling can even be counterproductive when corporations are found to be dishonest in their proposed support of sustainability. That has been the case in the context of data centers. Companies often focus on promoting the clean energy sources used by their data centers while covering up and ignoring other environmental harms resulting from their use. Google, for example, consumes enormous amounts of water in order to cool their data centers. Another revealing case comes from a now infamous incident in 2012 when Microsoft threatened to burn inordinate amounts of excess power in its data center to avoid paying a fine for overestimating consumption from the local power utility organization. These instances of major data center companies ignoring environmental considerations often undermines the power any signaling effect may have in moving towards sustainable initiatives.
Beyond the muddled impact of the signaling effect, the more direct harm from these data centers originates from their inefficient use of energy. Regardless of the source of energy, data centers often continue to operate at full capacity, even when it’s not needed from a business perspective. Data centers contributed about 0.3% of global carbon emissions in 2018, yet only about 10% of the energy used was actually needed for their operations. Given the 33.1 billion tons of carbon emissions in 2018, this translates to nearly 100 million tons of unneeded carbon emissions from data centers. Beyond the climate cost, the inefficiencies with how companies operate their data centers leads to a severe problem with e-waste. E-waste is rarely recycled and instead left in dumping grounds, largely in South and East Asia, causing severe harms to the local populace. One example comes from the Chinese city of Guiyu where studies have found 80% of children experience respiratory ailments due to the waste there. Further, as we explain more extensively below, there are many ways to mitigate the inefficiencies in this approach, but companies continue to ignore them.
Going back to our framework at the top, we can weigh the consequences of how companies operate these data centers. On one hand, the choice to move towards renewable sources of power for the data centers has led to some positive impact in terms of spurring more renewable energy implementations. On the other, the extent of this effect is mitigated by the hypocritical actions these companies take with their data centers in regards to other environmental concerns. The more direct negative consequences, to the environment can be hard quantify. One study from Oxford found that every extra metric ton of carbon emissions incurs a social cost of $42 due to harms to the environment. Using the excess carbon emissions we found above, this would mean data centers result in a $4 billion cost from energy usage not needed from their operations. These costs, alongside the harms to locals from waste and resource usage, are direct and foreseeable consequences of how data centers operate. From this we conclude that the inefficiencies with which companies run their data centers is immoral and thus we focus on the solutions and policy considerations of that below.
After seeing the environmental and ethical consequences of the status quo of cloud computing, it is clear that change is necessary. This change can come from either the companies who own and operate data centers or the government. From the data center side of the solution, the overall goal is reduce wasteful resource consumption. Two different strategies exist in order to accomplish this goal: algorithmic resource assignment and technological advancement.
As mentioned before, data centers waste megawatts of power each year. Globally, data centers consume about 30 billion watts of electricity, roughly equivalent to the output of 30 nuclear power plants, with the United States consuming between a quarter and a third of this amount. However, with such high consumption, 90% or more of the power consumed is only utilized for keeping the server computers powered, meaning data centers in the United States alone are wasting as much as 9 billion watts every year. The first solution to this problem is a better use of resources. Most data centers only run at 9-12% utilization, but the National Energy Research Scientific Computing Center at the Lawrence Berkeley National Laboratory in California was able to reach 96.4% utilization. While this was achieved by scheduling large computing jobs ahead of time, a strategy most large tech companies would be wary to rely on, significant research has been poured into the field of resource assignment as of late. This research can be applied to significantly shrink the size and number of data centers while still providing the same quality of service.
The first theory of resource allocation on which many others are based is Dominant Resource Fairness (DRF). DRF allocates users of a cloud computing service the resources for which their demand is highest, such as bandwidth, memory, and computer cores, and every other resource in proportion. While this theory improves on prior theories which attempted to package all of these resources into one, it still leaves many available resources underused. To improve upon DRF, Suhan Jiang and Jie Wu introduced 2-Dominant Resource Fairness (2DRF), an algorithm which expands the consideration of DRF to the two resources for which a user has the highest demand. Additionally, this algorithm can be expanded further to consider a number k resources to further improve efficiency. Another available algorithm is Dominant Resource with Bottlenecked Fairness (DRBF), an algorithm which expands upon DRF to organize users demands into different queues and divide resources evenly among these queues. These algorithms, 2DRF and DBRF, and many more like them could be used to reduce the number of servers required for a data center to operate efficiently, which would give companies which operate data centers the ability to eventually shrink the size of their data centers.
The second avenue for data centers to reduce their resource consumption is through technological advancements, specifically in their cooling. In an investigation of data centers in the southwest region, an area that is historically drought-ridden where careful water management is key, data centers often bully local water utilities with requests of 1.46 billion gallons of water a year. However, this water is used to cool servers in a particularly wasteful method of cooling: evaporative cooling. Rather than trying to reduce or eliminate water consumption, such as with a hybrid air-cooled chiller with free cooler (a method of cooling that requires no water), companies draw enough water for a small city and literally watch it disappear into thin air.
There are three primary areas in which public policy needs to progress:
The problem of real estate trusts doesn't really lie with the data centers themselves, it rather is a problem with America's tax code. By reorganizing themselves as real estate investment trusts (REIT), data center companies are able to get out of taxes, with one company getting out of more than 100 million in taxes annually.
A company in a capitalist society has an obligation to make as much money in profits as possible, so the data centers can't be faulted for wanting to reorganize. However, the government has an obligation to hold the corporations to their full tax burden. To this end, there is one primary course that the government ought to take: disallowing data centers from being classified as real estate investment trusts.
The government has this obligation because companies are given specific tax rates by democratically elected representatives, and, if through adminsitrative loopholes, companies manage to skirt their taxes, the administrative branch of the govenment has a duty to close that loophole.
Some companies ran their server farms with as low as 10% utilization of their total uptime during 2008 and the numbers haven't improved much in recent years, despite an abundance of optimization algorithms designed to help companies run their servers with much higher utilization.
With the looming threat of climate change, this wasted energy usage is an outrage. Companies not only have a categorical imperative to protect the Earth as inhabitants of it, but additionally there are utilitarian reasons why companies should waste less energy (lower energy bills, prolonging the lifespans of their consumers, protecting their investment in real estate, etc.).
With these ethical reasons why the companies should not behave in the way they are, a casual observer is left with no other assumption that the companies are behaving unethically by managing their servers in the current fashion. Because of this, governments at some level must intervene to create minimum efficiency standards for these companies. The standards need not be too drastic even. Mandating a standard as low as 50% utilization would allow companies to maintain high levels of service uptime, and would lower the total wasted energy by a factor of more than 500% in some cases.
This goes hand in hand with the previous point, but many companies can and would absorb the cost of a fine rather than comply with legal restrictions because the penalties for violating energy usage laws are not thorough enough to ensure compliance. In one example, Microsoft incurred a $210,000 penalty for drastically overestimating its energy usage. It could, rather than pay the penalty, increase its usage wastefully by $70,000 to avoid the penalty.
Though that specific situation may not have been fixed by larger penalties (in fact, it would have exacerbated the problem), more thorough regulations including minimum efficiency standards would have prevented the problem. To close loopholes, local, state, and federal governments would have to cooperate on a large scale, but such is the necessity of an emergency like climate change. An example plan might include: