Using AI to mitigate the growing environmental threat of data centers

Christina Delimitrou, a newly tenured associate professor at MIT, is addressing the environmental strain caused by the global building boom of power-hungry data centers. She works in the Department of Electrical Engineering and Computer Science and is a member of the Computer Science and Artificial Intelligence Laboratory.

Delimitrou and her group apply machine learning to make large-scale data centers more efficient, secure, and reliable. They redesign outdated cloud computing systems, develop methods to manage shared hardware resources, and create streamlined server architectures, allowing operators to get more computational power from existing hardware. She notes that data centers not utilized to their full capabilities burn more power than needed, and removing software "bloating" without compromising performance could reduce the need to build new data centers. Her work also uses AI to help programmers find and fix problems in cloud-based applications, eliminating downtime.

During graduate study at Stanford University with mentor Christos Kozyrakis, Delimitrou found that many large computing systems ran at only about 15 percent capacity rather than near 100 percent. To push utilization closer to full, she investigated machine-learning solutions to automate resource management. After earning her PhD, she continued this work as an assistant professor at Cornell University, where her group developed Seer, a deep-learning tool that anticipates and prevents problems in web applications. She joined MIT as an assistant professor in EECS in 2022.

At MIT, Delimitrou extended her debugging work to cover security issues that can make user data vulnerable to hackers. She also uses AI to redesign software systems to better fit existing hardware, and is working on adding explainability to AI tools so people can get useful feedback. Because companies operating data centers use proprietary hardware and inaccessible software, her group creates clones of proprietary systems; one tool, Ditto, mimics an application's structure and performance characteristics.

Why it matters: These advances could let data center operators extract more computational power from existing hardware, potentially reducing the need to build additional data centers and lessening reliance on fossil-fuel energy. Improved resource management could also give end users more predictable performance from applications on their smartphones.