Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Improve partition_load_balance #2206

Merged
merged 15 commits into from
Feb 27, 2024

Conversation

thorstenhater
Copy link
Contributor

Spin-off from #2005. Make the primary load balancing cleaner and faster and more maintainable.
Thus:

  • remove all MPI calls, this is now purely local
  • remove temporary data structures and/or coral them into their own little scopes
  • simplify super_cells vs regular_cells
  • sort less
  • sparse connection tables

Partially inspired by external feedback

- Remove redundant MPI call for global gid list
- Make GJ table local
- Extract connection table builing.
- add group parameters struct to bundle info
- coral temporary structures into their own scopes to avoid RSS growth.
Also, add weird test to see that we can _almost_ construct the partition.
example/busyring/init-only-2048-complex.json Outdated Show resolved Hide resolved
example/busyring/ring.cpp Outdated Show resolved Hide resolved
arbor/partition_load_balance.cpp Outdated Show resolved Hide resolved
arbor/include/arbor/common_types.hpp Show resolved Hide resolved
arbor/partition_load_balance.cpp Show resolved Hide resolved
arbor/partition_load_balance.cpp Show resolved Hide resolved
test/unit/test_domain_decomposition.cpp Show resolved Hide resolved
ctx->distributed = std::make_shared<distributed_context>(dummy_context{rank, nranks});
for (const auto& R: {gj_symmetric(nranks, true), gj_symmetric(nranks, false)}) {
// NOTE: This is a bit silly, but allows us to test _most_ of
// the invariants without proper MPI support. If we could get `gather_gids` to
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

just curious: can you think of a way to make gather_gids work?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Not really without major surgery. We could split this function into three, tentatively:

  1. register groups, called in parallel and shoving the data into the decomposition
  2. gather_gids, collective call
  3. actually construct the decomposition

but that'll require putting the data threaded between those calls into the decomposition
since they must be independent and pass around some state.

@thorstenhater
Copy link
Contributor Author

@boeschf the failure seems to be a timeout on CSCS CI/CD?!

@thorstenhater
Copy link
Contributor Author

@boeschf any news here?

Copy link
Contributor

@boeschf boeschf left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

sorry it took so long! looks good!

@thorstenhater thorstenhater merged commit ed67763 into arbor-sim:master Feb 27, 2024
19 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

2 participants