Ask the Experts – How Are Data Centers Impacted by Climate Change?
BY TRISHA BALAN & GINA MANTICA
Our digital lives are often removed from the physical world, but with the growing problem of climate change, the physical is beginning to adversely affect the digital. Data centers, the key facilities that house computer systems, are becoming less efficient with changes in global temperatures.
Over the summer, the United Kingdom (UK) and the United States experienced a large number of data center shutdowns due to overheating caused by extremely hot weather. In July, the UK heatwave caused Google and Oracle’s data centers to shut down. In September, Twitter’s California data center experienced a major outage due to the intense heat. Rising global temperatures and unstable weather fluctuations have made maintaining data centers a hard task. This costs companies millions, negatively affects user experience, and contributes to climate change through extreme energy consumption. In light of this problem, scientists and corporations are looking for new ways to create more energy-efficient cooling systems to support digital infrastructures.
We asked two energy-efficient computing experts, Ayse Coskun, Director of the Center for Information and Systems Engineering (CISE), and Daniel Wilson, a graduate researcher with BU’s Performance and Energy Aware Computing Lab (PeacLab), about the effects of climate change on maintaining data centers and the future of energy-aware technology.
How do data centers deal with temperature fluctuations?
If the external temperature fluctuates (e.g., a hotter summer day than usual), cooling systems, especially those related to cooling the data center building. “work harder” to maintain safe room temperatures. Working harder means typically spending more energy (and therefore money) to cool down. Or in more extreme cases, when working harder to cool down the data center doesn’t cut it, it may be necessary to shut down or throttle part of the data center, which then impacts the performance and overall efficiency of the data center.
What happens if the center overheats?
If a data center overheats, it either needs to generate less heat or move more heat away from the data center. Generating less heat may mean performing less computing work, which could make users of the data center less satisfied with their service. Moving more heat away from the data center might mean turning on more cooling systems (such as air chillers), which would increase the power draw (and operational cost) of the data center.
If a data center fails to address overheating issues, it may reduce its components’ lifespan and computing performance, causing both dissatisfied users and increased operational costs.
What technology exists to keep data centers cool?
Many factors are considered when developing cooling systems for data centers like low-level chip and heat sink design, thermally-aware software, the physical arrangement of servers, fans, and a building’s cooling systems, and site design decisions that exploit a data center’s local climate.
Conventional cooling designs include heat sinks, fans, liquid-cooling systems, and compression-based chillers. They figure out how to safely operate the data center at higher temperatures, how to arrange their servers to achieve optimal airflow, and how to exploit local climate features more effectively (like when you open your windows in the Fall instead of running an air conditioner). More extreme solutions are also under investigation, such as Microsoft’s Project Natick, which moves data centers undersea for cooling and other benefits.
How can data centers prepare for future warming?
An ACM Tech Brief from last year explains that in order to meet climate change recommendations from the 2015 Paris Agreement, we need to globally achieve net-zero emissions by the year 2050. At the time of the report, data centers were estimated to consume 3% of the global energy supply, and that share is increasing. In other words, it will not be sufficient for data centers to simply prepare for the temperature impacts of global warming, but data centers must also actively work toward reducing contributions to global warming.
Data centers and researchers are working toward that goal in several ways. Some data centers work toward increasing their efficiency (how much work they complete with a given amount of energy), cooperating in a smart grid to use less energy when the power supply has a high carbon impact, relocating their work to follow green energy supply, or even using their own renewable energy sources.
How does your research mitigate the impacts of climate change on the digital world?
Our recent work in PeacLab investigates how a data center can participate in “demand response” programs. “Demand response” programs give an electricity user monetary credit in exchange for allowing the electricity provider to influence the user’s power demand. As a result, the electricity provider can handle mismatches between supply and demand by changing the demand instead of changing the supply, which may require costly or eco-unfriendly actions.
For example, a participating data center may be asked to increase its power consumption at one time, then it may be asked to decrease its power consumption a few seconds later. Our work determines how much power a data center should offer to help balance supply and demand. Our work also designs methods to honor data center demand response commitments while meeting the computational performance needs of the data center’s users.
Our lab also works across different levels of the “computing system stack” (from hardware architecture to operating systems and applications) to improve thermal and energy efficiency of computer systems. For example, we design optimizers that improve processor thermal and energy efficiency (e.g., by tuning the architecture or placement of components on the chip), while adhering to the performance constraints such as delivering a certain amount of throughput per second.
We design simulators that enable evaluating thermal and energy efficiency of new processor technologies to easily test future designs for more efficient and reliable systems. We also develop policies at a single processor level or across a larger system like a data center to manage temperature and efficiency at runtime under the dynamic workload and environmental conditions. All these methods collectively work towards a more sustainable computing future.
*Responses have been edited and condensed for clarity.