Skip to content
This repository has been archived by the owner on May 20, 2024. It is now read-only.

Where should we be putting parallelism? #41

Open
cathieO opened this issue Jul 16, 2019 · 2 comments
Open

Where should we be putting parallelism? #41

cathieO opened this issue Jul 16, 2019 · 2 comments
Labels

Comments

@cathieO
Copy link

cathieO commented Jul 16, 2019

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

@cathieO cathieO added the Epic label Jul 16, 2019
@cathieO
Copy link
Author

cathieO commented Jul 16, 2019

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

@cathieO
Copy link
Author

cathieO commented Sep 6, 2019

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.

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
Projects
None yet
Development

No branches or pull requests

1 participant