Facebook is committed to sustainability and the fight against climate change. For this reason, we announced in September 2020 that we would achieve net zero emissions in our value chain by 2030. Part of Facebook’s sustainability efforts includes data center efficiency – from building servers that use less energy to run, to developing a liquid cooling system that uses less water.
To find out more about these sustainability efforts in the data center on Facebook and how we are working with the academic community in this area, we contacted Dharmesh Jani (“DJ”), the Open Ecosystem Lead in the Hardware Engineering Team and Open Compute, composed of IC Chair and Dr. Katharine Schmidtke, Director of Sourcing at Facebook for application-specific integrated circuits and customer-specific silicon. The teams from DJ and Schmidtke are working to achieve four goals:
- Extend the life cycle of Facebook data center devices and make our devices reusable for others.
- Improving the energy efficiency in the Facebook infrastructure through hardware and software innovations.
- Reduce the carbon-intensive content in our data centers.
- Work together with industry and science to drive innovation in terms of sustainability in our value chain.
DJ and Schmidtke discuss why it is important to build data centers that are as energy efficient as possible, how we work with and support scientists and industrial partners in this area, and what potential research challenges Facebook researchers and engineers could address next.
Establishment of energy-efficient data centers
Facebook has been committed to sustainability and energy efficiency for over a decade. In 2009, Facebook built its first in-house data center in Prineville, Oregon, one of the most energy-efficient data centers in the world with a PUE (Power Usage Effectiveness) ratio between 1.06 and 1.08. In 2011, Facebook shared its designs with the public and, along with other industry experts, launched the Open Compute Project (OCP), a rapidly growing global community whose mission is to create, use, and enable the most efficient drafts for the design Computing. But there is more to be done.
“Data centers use an average of 205 TWh of electricity per year, which corresponds to 145 tons of CO2 emissions,” explains DJ. “As hyperscale data centers grow in the coming years, this emission will increase dramatically if the mitigation is not taken into account today (Source 1, Source 2). Facebook also wants to work on this growing emission to make sure we are efficient and meet our goal of net carbon zero by 2030. “
You may also like
Chasing Carbon: The Elusive Computer Footprint
According to DJ, Facebook is taking several steps to address these issues: “Facebook’s sustainability team is working across organizations to meet the goals that will lead to a reduction in carbon emissions. Circularity is one of the emerging efforts within the infrastructure to extend the life cycle of devices, which has the greatest impact on net zero carbon expenditure. With the Open Compute project, we are driving sustainability and circularity efforts in the industry, ”he says.
Building a data center itself also contributes to carbon emissions. A high utilization efficiency in already built data centers is the key to reducing the demand for the construction of new data centers. Over the years, Facebook has developed a number of industry-leading technologies to control and manage the peak power needs of data centers. As a result, many more servers can be hosted in existing data centers with limited power capacity. This has reduced the demand for data center construction by more than 50%. The technology is developed in-house through academic collaborations and research internship programs. Some key research and high-level industrial operations experiences are also shared with the community through high-level publications at academic conferences. Here are a few examples: Dynamo: Facebook’s data center-wide energy management system, coordinated, priority-conscious charging of distributed batteries in oversubscribed data centers.
Find out more about the efficiency of Facebook data centers in the Tech @ blog and read our latest sustainability report in the newsroom.
Partnerships and collaborations
Industry cannot develop energy-efficient technologies on its own. That is why we often work with academic experts and support their pioneering work. “In recent years, Facebook has started a series of research collaborations on the subject of energy saving and energy efficiency,” says Schmidtke. “Recently, Facebook sponsored the Energy Efficiency Institute at UC Santa Barbara with a $ 1.5 million gift over three years. We hope that our contribution will help advance research on the energy efficiency of data centers. “
“Another example is the ongoing research collaboration with Professor Clint Schow at UCSB,” says Schmidtke. “The project focuses on increasing the efficiency of optical data transmission between servers in our data center network. The research has just entered the second phase and aims at highly efficient coherent optical links for data transmission. “
Facebook is also an industry member of the Center for Energy-Smart Electronic Systems (in partnership with the University of Texas at Arlington) and the Future Renewable Electric Energy Delivery and Management Systems Engineering Research Center (at North Carolina State University).
Facebook not only promotes innovation in the academic community, but also uses industry partners. DJ: “We want to drive initiatives in the field of sustainability within the OCP community in order to align other players in the industry along the value chain. We plan to define sustainability as one of the principles of the OCP so that all future contributions can focus on it. “
What’s next
DJ offers three sustainability challenges that researchers in the field could address next. All of this would involve industry collaboration with universities and other research organizations.
One research challenge is to make the calculation more climate-neutral. The computing requirements of the AI field have grown exponentially: Since 2012, the computing effort for the largest AI training runs has increased exponentially with a doubling time of 3.4 months – a 300,000-fold increase in computing power from AlexNet to AlphaGo Zero. “How can we make AI more efficient when the current approach to increasing computation is impractical?” says DJ. “This is one of the biggest challenges in this area. I look forward to further green AI initiatives.”
Another challenge is planning workloads (WL) in data centers when carbon intensity is low. “We need to think about the amount of WL that is going into the data centers and complex interactions to optimize them for such use cases,” explains DJ. “I hope to find new algorithmic ways to reduce energy consumption, distribute the workload and influence carbon emissions.”
Another potential focus is the technology that uses chiplets. Chiplets can be viewed as reusable mix-and-match building blocks that together form more complex chips. This is a more efficient system that uses a smaller carbon footprint. “I’m looking forward to new computer architectures that are domain-specific and controlled by chiplets,” says DJ. “We’ve only explored the tip of the iceberg in terms of sustainability. In this area we can do a lot together to achieve the goal of a greener future. “
Facebook is committed to open science and we value our partnerships with industry and science. We are confident that together we can help drive technology and innovation in this area.
Comments are closed.