ExaDigiT: Digital Twin for Exascale Supercomputers (2022 - Present)

Overview

A digital twin is a virtual replica of an HPC system, enabling scenario modeling and optimization. This project integrates real-time telemetry with physics-based models to optimize energy efficiency and system stability.

• Co-led the design of ExaDigiT, a digital twin for Frontier’s liquid cooling system, enhancing predictive control strategies.
• Developed a real-time data integration layer linking system telemetry with digital twins.
• Published initial findings at SC24 and contributed to a global digital twin initiative for HPC centers.

Technology

• Python REST API based data access towards Apache Druid
• Kafka based simulation task dispatch and management
• Modelica based cooling models
• Python based scheduler simulation
• Unreal engine for AR/VR (Microsoft Hololens)

Publication

• Brewer et al., “A Digital Twin Framework for Liquid-cooled Supercomputers as Demonstrated at Exascale”, SC24, 2024
• Shin et al., “Navigating Exascale Operational Data Analytics: From Inundation to Insight”, SusSup24 - SC24-W, 2024