MIT News breed · Science

AI to Enhance Data Center Efficiency

Christina Delimitrou at MIT uses AI to make data centers more efficient, addressing their growing environmental impact from energy consumption.

Her research applies machine learning to redesign cloud systems and manage hardware, enabling data centers to extract more power from existing equipment. This reduces the need for new construction and reliance on fossil fuels.

AI also helps programmers fix application issues, preventing downtime that wastes resources and ensuring predictable performance for users.

Delimitrou's early interest in math and engineering led her to study computer engineering, focusing on resource management.

At Stanford, she found many large computing systems operated at only about 15 percent capacity, highlighting inefficiency.

She investigated machine learning for automated resource management, a novel approach at the time. Tools like Seer use deep learning to prevent application problems.

At MIT, her team adds explainability to AI tools for better system design insights. They create system clones to study proprietary environments.

Delimitrou stresses careful AI application and auditing, ensuring efficiency and understandable insights for future designs, acknowledging limitations with proprietary systems.

AI-samenvatting op basis van de bron.