Performance modeling

Predicting large-scale simulation performance and energy use

Problem

Running large biomedical simulations without performance foresight can waste allocation time and energy. I investigated whether interpretable analytical models can accurately predict runtime behavior across hardware and load-balance scenarios.

Approach

  • Derive per-process symbolic performance models from code-level execution structure.
  • Calibrate model parameters with targeted empirical measurements.
  • Validate predictions across balanced and imbalanced execution scenarios.
  • Compare behavior on multiple large-scale compute architectures.

Key finding

The calibrated analytical model achieved strong prediction quality across tested cases while retaining interpretable bottleneck structure.

Why it matters

Accurate, understandable performance forecasts improve experiment planning, hardware migration decisions, and energy-aware simulation strategy.

Outputs

  • Publication details are listed in the References section below.
  • Model structure and validation examples are summarized on this page.

References

2022

  1. Building a Fine-Grained Analytical Performance Model for Complex Scientific Simulations
    Jelle Dijk, Gabor Zavodszky, Ana-Lucia Varbanescu, and 2 more authors
    In International Conference on Parallel Processing and Applied Mathematics, 2022