Fusing Experiments, Data, and Computing at DIII-D and ALCF

Mark Kostuk
Brian Sammuli
Webinar

– CT

Virtual
Register
alf_ses graphic featuring title and speakers.

Modern experimental user facilities increasingly rely on a tight integration between experimental operations, high-performance computing (HPC), and streamlined data access. This two-part presentation highlights how the DIII-D National Fusion Facility is pioneering this space through automated remote analysis and the Fusion Data Platform (FDP), serving as a blueprint for the broader Integrated Research Infrastructure (IRI) initiatives.

Part 1: Automatic On-Demand Remote Analyses for Fusion Experiments: Mark Kostuk, General Atomics

This section will focus on the DIII-D "digital twin," demonstrating how automated, on-demand workflows utilize the Globus API (Flows and Compute) to trigger resource-intensive plasma modeling codes at leadership compute facilities (LCFs) like NERSC and ALCF immediately following an experiment.

Part 2: An AI-Ready Data Harness for Fusion: The Fusion Data Platform: Brian Sammuli, General Atomics

This section introduces the Fusion Data Platform (FDP), a software stack that delivers fusion data to any machine in the world through one common interface. FDP offers reusable patterns for scientists to reach and combine data across devices, integrating leadership-class facilities and AI workflows.

Speakers

Dr. Mark Kostuk obtained his PhD in physics from The University of California, San Diego with a focus on nonlinear dynamics system identification, control theory and chaotic data assimilation. Since then he has been applying these skills to fusion simulation, modeling and data analysis at General Atomics and the DIII-D National Fusion Facility. While there, Kostuk led a group of collaborators to be among the first to address the challenge of on-demand, remote execution of large, high-fidelity simulations at a leadership compute facility in support of ongoing plasma experiments at DIII-D. He currently leads the DIII-D digital twin development effort, and is presently focused on the problems of heterogenous model integration, modularity, and performance-at-variable-scales that lie at the core of the digital twin challenge.

Brian Sammuli is the Lead for Applied Machine Learning and the Deputy Director of the Advanced Computing Center within the Energy Group at General Atomics. He leads the development of AI, machine learning, and data technologies for fusion energy, serving as Principal Investigator for the DOE-funded Fusion Data Platform and leading multiple initiatives in plasma control, scientific machine learning, and large-scale fusion data infrastructure. His work spans real-time plasma control, digital twins, distributed scientific data systems, and AI-driven fusion research, including the development of software and infrastructure used to accelerate fusion experimentation and enable next-generation fusion energy systems.

Register