Application of the extreme scaling computing pattern on multiscale fusion plasma modelling

Area of Science:

• Computational physics
• Plasma physics
• High-performance computing

Background:

• Magnetic confinement fusion research relies on complex simulations.
• Multi-scale workflows are essential for modeling plasma behavior across different timescales.
• The ComPat project's scaling approach is adapted for fusion simulations.

Purpose of the Study:

• To apply the ComPat project's extreme scaling pattern to a multi-scale workflow for magnetic confinement fusion.
• To investigate the impact of performance metrics on optimal execution configurations.
• To present and evaluate a more parallelized workflow variant.

Main Methods:

• Integration of transport, turbulence, and equilibrium codes into a multi-scale workflow.
• Analysis of performance metrics including time to completion, efficiency, and energy consumption.
• Development and testing of a variant workflow with increased inherent parallelism.

Main Results:

• Initial profile measurements indicate that optimal core allocation depends on the chosen performance metric.
• The new workflow variant achieves equivalent simulation results.
• The parallelized variant typically demonstrates lower computational cost compared to the original workflow.

Conclusions:

• The choice of performance metric significantly influences the optimal execution configuration for multi-scale fusion simulations.
• A more parallelized workflow variant can achieve comparable results with reduced computational cost.
• This work contributes to efficient exascale computing for fusion energy research.