You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Mesh migration can be rather slow on large distributed meshes because each node is sent and received individually with blocking directives. I don't have a large file at the moment but on a moderate-size acoustics problem run on 64 Frontera nodes the timing was:
migration time: 4.14451 seconds
This becomes increasingly expensive on large meshes.
Indirect communication and/or batching may be needed.
The text was updated successfully, but these errors were encountered:
jbadger95
changed the title
Indirect communication in mesh migration
Blocking communication in mesh migration
Mar 10, 2023
jbadger95
changed the title
Blocking communication in mesh migration
Blocking directives in mesh migration
Mar 10, 2023
As an additional comment: at the moment, a lot of node data (except for middle nodes) are sent/received redundantly; there are various ways of avoiding this which should help with the performance of the data migration.
Mesh migration can be rather slow on large distributed meshes because each node is sent and received individually with blocking directives. I don't have a large file at the moment but on a moderate-size acoustics problem run on 64 Frontera nodes the timing was:
This becomes increasingly expensive on large meshes.
Indirect communication and/or batching may be needed.
The text was updated successfully, but these errors were encountered: