The fast growth of data centers is putting pressure on electrical grids. This has led to greater use of energy from fossil fuels, according to MIT News.
Christina Delimitrou, a newly tenured associate professor at MIT, is working on this problem. She and her group use machine learning to make large data centers more efficient, secure, and reliable.
Their work includes redesigning outdated cloud computing systems and managing shared hardware. The goal is to get more computing power out of hardware that already exists.
"If data centers are not utilized to the best of their capabilities, then they will burn much more power than they need to meet growing user demand," Delimitrou said. She added that removing software "bloating" without hurting performance could reduce the need for new data centers.
Finding Unused Capacity in Data Centers
Delimitrou grew up in northern Greece and studied computer engineering at the National Technical University of Athens. Her final-year thesis looked at managing resources when several applications run on one computer at once.
She then went to Stanford University for graduate school. There, she and her mentor, Christos Kozyrakis, studied inefficiencies in cloud systems.
They found that most large computing systems were running at only about 15 percent capacity. "This is not a resource-efficient or sustainable way of scaling these systems," she said.
To raise that number, she turned to machine learning. She said the approach was risky at the time because no one had yet shown it would work.
After her PhD, she became an assistant professor at Cornell University. Her group there built Seer, a tool that uses deep learning to predict and prevent problems in web applications before they cause slowdowns.
She later noticed developers were splitting applications into smaller pieces spread across many servers. Because servers were not built for this design, she reworked parts of her earlier research to handle it.
AI Tools for Cloud Systems at MIT
Delimitrou joined MIT's Department of Electrical Engineering and Computer Science as an assistant professor in 2022. She is also a member of the Computer Science and Artificial Intelligence Laboratory.
Her team uses AI to help programmers find and fix problems in cloud apps, such as music streaming and video calls. She has expanded this work to include security flaws that could expose user data.
She said studying these systems has become harder because tech companies use private hardware and software. To work around this, her group creates clones of these systems.
One clone tool, called Ditto, copies an application's structure and performance. This allows researchers to run a wide range of studies.
Her team is also adding explainability to its AI tools. "One of the challenges when it comes to applying AI to these systems is that the AI is not interpretable," she said.
Delimitrou expects her work to keep changing as machine learning models improve. "We still need to audit it and be especially careful about how these models are applied," she said.
