Using AI to mitigate the growing environmental threat of data centers
By rethinking how large cloud computing systems operate, Associate Professor Christina Delimitrou seeks to make data centers more energy efficient.
The rapid expansion of data centers is putting a strain on electrical grids and increasing reliance on fossil fuel energy sources. To address this environmental issue, Christina Delimitrou, a tenured associate professor at MIT, is leveraging machine learning to enhance the efficiency, security, and reliability of data centers. By reimagining outdated cloud computing systems, optimizing shared hardware resources, and refining server architectures, Delimitrou and her team enable data center operators to extract more computational power from existing hardware.
According to Delimitrou, data centers often underutilize their resources, which results in excessive power consumption to meet growing user demand. By eliminating unnecessary software bloat in a manner that does not compromise performance, the need for constructing new data centers can be reduced. Furthermore, Delimitrou harnesses AI to assist programmers in identifying and rectifying issues in cloud-based applications, such as music streaming services or video conferencing platforms, thereby minimizing application downtime that consumes computational resources.
Delimitrou's interest in math and science originated from her upbringing in Greece, where she was exposed to ancient history and encouraged by her parents, a chemical engineer and a pharmacist. After studying computer engineering at the National Technical University of Athens, she pursued her graduate studies at Stanford University, focusing on inefficiencies within cloud computing systems and large-scale data centers.
Delimitrou's research revealed that many large computing systems operate at only 15% capacity, which is neither resource-efficient nor sustainable.
By applying machine learning to streamline computational processes, Delimitrou's work aims to push utilization rates closer to 100%. This approach automates resource management operations in the cloud, identifying solutions that developers might overlook. One such tool, Seer, utilizes deep learning to predict and prevent problems in web applications before they occur, averting potential slowdowns.
Delimitrou's research also explores new ways to address the evolving nature of cloud applications, including server designs for split applications.
Written by urgent.news from MIT News AI's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
