Dynas+

Case studies

Optimizing parallel computing for SPH simulations

1 min read
Industry
Aeronautics
Discipline
SPH & HPC
Technologies
LS-DYNA · LS-PrePost · LS-OPT
In partnership with
Safran Aircraft Engines
Parallel computing optimization for SPH simulations

The case

Simulations using the SPH (Smoothed Particle Hydrodynamics) method make it possible to reproduce strongly nonlinear phenomena such as hypervelocity impacts or bird strikes. In return, these models can contain several hundred thousand, or even several million, particles, resulting in very long computation times. Dynas+ developed and validated a methodology for optimizing the MPP (Massively Parallel Processing) decompositions of LS-DYNA in order to improve computational efficiency on multi-core architectures, in partnership with Safran Aircraft Engines.

The solution deployed

Dynas+ studied different MPP decomposition strategies applied to several representative industrial cases:

  • hypervelocity impact of a projectile on a plate modeled entirely in SPH;
  • Bird Strike simulation on fan blades with SPH / shell element coupling;
  • Bird Strike simulation on blades modeled with solid elements.

For each of these cases, several strategies for distributing the computation domains were developed and compared in order to:

  • limit communications between processors;
  • improve load balancing;
  • significantly reduce computation times without altering the physical results of the simulation.
Parallel computing optimization for SPH simulations, the solution deployed

The results achieved

The work made it possible to establish practical recommendations for optimizing MPP computations in LS-DYNA according to the nature of the SPH models. The best decomposition strategies deliver in particular:

  • a reduction of more than 50% in computation time on coupled SPH / finite element models;
  • improved scaling of parallel computations on multi-core architectures thanks to partitioning strategies tailored to SPH models.

This work was the subject of a scientific paper presented at the 13th International LS-DYNA Users Conference (Detroit): "Advanced MPP Decomposition of a SPH Model" (Vincent Lapoujade, Christian Mouteliere, Antoine Millecamps).

Parallel computing optimization for SPH simulations, the results achieved