Benchmarks

The following benchmarks represent typical use cases of YMeRo. They were performed on the Piz-Daint supercomputer for both strong and weak scaling. See in benchmarks/cases/ for more informations about the run scripts.

Bulk Solvent

Periodic Poiseuille flow in a periodic domain in every direction, with solvent only. Timings are based on the average time-step wall time, measured from the SimulationStats plugin.

../_images/strong_solvent.png

strong scaling for multiple domain sizes

../_images/weak_solvent.png

weak scaling efficiency for multiple subdomain sizes

The weak scaling efficiency is very close to 1 thanks to the almost perfect overlapping of communication and computation.

Bulk Blood

Periodic Poiseuille flow for blood with 45% Hematocrite in a periodic domain in every direction. Timings are based on the average time-step wall time, measured from the SimulationStats plugin.

../_images/strong_blood.png

strong scaling for multiple domain sizes

../_images/weak_blood.png

weak scaling efficiency for multiple subdomain sizes

The weak scaling efficiency is lower than in the solvent only case because of the complexity of the problem:

  • Multiple solvents
  • FSI interactions
  • contact interactions
  • Many objects
  • Bounce back on membranes

The above induces a lot more communication than the simple solvent only case.

Poiseuille Flow

Poiseuille flow between two plates (walls), with solvent only. Timings are based on the average time-step wall time, measured from the SimulationStats plugin.

../_images/strong_walls.png

strong scaling for multiple domain sizes

../_images/weak_walls.png

weak scaling efficiency for multiple subdomain sizes

Rigid Objects suspension

Periodic Poiseuille flow for rigid suspensions in a periodic domain. Timings are based on the average time-step wall time, measured from the SimulationStats plugin.

../_images/strong_rigids.png

strong scaling for multiple domain sizes

../_images/weak_rigids.png

weak scaling efficiency for multiple subdomain sizes