Performance analysis of electronic structure codes on HPC systems: a case study of SIESTA

Area of Science:

• Computational physics
• Materials science
• Quantum chemistry

Background:

• Accurate simulations of molecular dynamics require efficient electronic structure codes.
• Scaling performance of these codes on supercomputers is crucial for large-scale simulations.
• SIESTA is a widely used code for ab initio molecular dynamics.

Purpose of the Study:

• To evaluate the scaling and timing performance of the SIESTA code.
• To provide guidance for optimizing SIESTA execution on diverse supercomputing architectures.
• To propose a method for quantifying strong scaling efficiency.

Main Methods:

• Systematic weak and strong scaling tests were conducted.
• Tests were performed on six PRACE Tier-0 supercomputers with varied architectures.
• A novel measure for strong scaling efficiency was developed and applied.

Main Results:

• Efficiency increased with simulation size across all tested architectures.
• Scaling behavior differed qualitatively based on supercomputer topology.
• Absolute timings revealed optimal system sizes and core counts for each machine.

Conclusions:

• The SIESTA code demonstrates scalable performance on large-scale supercomputers.
• Supercomputer architecture significantly influences scaling efficiency.
• Results offer valuable insights for users and developers of electronic structure codes.