This week Christina Delimitrou, a CSAIL principal investigator and assistant professor in MIT’s Electrical Engineering and Computer Science Department (EECS), received a Presidential Early Career Award for Scientists and Engineers (PECASE) from the U.S. government via the National Science Foundation. She received the honor for her early-career work on improving the efficiency of large-scale datacenters.
This past Wednesday, President Joe Biden announced Delimitrou’s honor amongst nearly 400 other recipients. Established in 1996, the PECASE recognizes scientists and engineers who “show exceptional potential for leadership early in their research careers.”
Delimitrou focuses on improving the performance capabilities of large-scale datacenters, helping them host tens of thousands of diverse applications each day more efficiently. To accomplish this, she’s developed methods that schedule applications in a way that makes the best use of the servers while preserving the application’s quality of service (QoS) requirements.
She also focuses on exploring how hardware design can help support new cloud programming frameworks like microservices and serverless. This includes designing efficient server architectures, distributed performance debugging, and cloud security techniques.
Delimitrou’s recent work illustrates her aims to enhance datacenter efficiency. Last year, her team helped develop “Sabre,” an approach that takes advantage of near-memory analytics accelerators commonly used in modern datacenter processors to restore computer memory faster. This method helped the researchers build a prefetching framework — a collection of pre-loaded resources likely to be used in a cloud application — that can quickly retrieve memory from snapshots (a previous saved state of the system), reducing performance overheads and using available resources more efficiently.
She and her colleagues also found flaws in the machine-learning models used to ensure the balanced use of resources across the global network of computers, servers, and datacenters Google uses to run services like Gmail and YouTube. These AI systems, which are used in cloud environments, struggle to generalize to out-of-distribution (unfamiliar) data. The team presented these findings in a paper selected as one of the Best of Computer Architecture Letters (CAL) for 2024.
The Presidential Early Career Award is the latest of many honors in Delimitrou’s burgeoning career. She previously received a Sloan Faculty Research Award, two Google Faculty Research Awards (2019 and 2020), a Microsoft Research Faculty Fellowship, the 2020 IEEE TCCA Young Computer Architect Award, an Intel Rising Star Award, a Google Research Award in Recognition of Technical Leadership and Achievements in Systems Research, a Facebook Faculty Research Award, the Cornell Excellence in Research Award and Excellence in Teaching Award, and several best paper honors.
