Load-balancing

Dynamic load-balancing of large-scale cellular simulations

Problem

Large-scale cell-based blood simulations naturally produce uneven particle distributions, which can degrade parallel efficiency. I studied how much of total runtime overhead is attributable to load imbalance versus communication.

Approach

  • Define a fractional-overhead model for load imbalance in particle-rich flow simulations.
  • Measure imbalance overhead in parallel lattice Boltzmann blood simulations.
  • Compare measured imbalance against communication overhead across problem scales.
  • Validate model predictions against observed performance data.

Key finding

The analytical model matched measured overhead trends well. Communication dominated in tested configurations, but the model predicts imbalance becomes increasingly dominant as systems grow.

Why it matters

Reliable imbalance prediction enables proactive runtime control and better scheduling decisions for high-cost biomedical HPC workloads.

Outputs

  • Publication details are listed in the References section below.
  • Performance diagnostics and distribution visualizations are shown on this page.

References

2019

  1. Optimizing parallel performance of the cell based blood flow simulation software HemoCell
    Victor Azizi Tarksalooyeh, Gábor Závodszky, and Alfons G Hoekstra
    In Computational Science–ICCS 2019: 19th International Conference, Faro, Portugal, June 12–14, 2019, Proceedings, Part III 19, 2019

2018

  1. Load balancing of parallel cell-based blood flow simulations
    Saad Alowayyed, Gábor Závodszky, Victor Azizi, and 1 more author
    Journal of computational science, 2018