Where should we be putting parallelism?

hydroframe / ParFlow_PerfTeam

Parflow is an open-source parallel watershed flow model.

http://inside.mines.edu/~rmaxwell/maxwell_software.shtml

Other

0 stars 0 forks source link

Where should we be putting parallelism? #41

Open cathieO opened 5 years ago

cathieO commented 5 years ago

Should we be calling parallel hypre? This will depend on profiling data.

cathieO commented 5 years ago

PFMG is called 9100 times 15% of 4 minutes or so. This can be investigated with the benchmark suite.

cathieO commented 5 years ago

Make a decison on parallelism for the full domain. The box sizes depend on the domain. There is a partitioning algorithm, clustering. Berger-Rigoutsos Seems to be tunable. Balance having some boxes with inactive cells. Can we say what the smallest box is?

We have examples that are a single layer. We have a histogram of box sizes, but they aren't representative.

We don't have the dataset for the rectilinear space.