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

coll/tuned: Change the default collective algorithm selection #7952

Merged
merged 1 commit into from
Aug 3, 2020

Conversation

wckzhang
Copy link
Contributor

@wckzhang wckzhang commented Jul 20, 2020

The default algorithm selections were out of date and not performing
well. After gathering data from OMPI developers, new default algorithm
decisions were selected for:

allgather
allgatherv
allreduce
alltoall
alltoallv
barrier
bcast
gather
reduce
reduce_scatter_block
reduce_scatter
scatter

Signed-off-by: William Zhang wilzhang@amazon.com
(cherry picked from commit ce40cfb)

@jsquyres
Copy link
Member

These changes need to go in on master before they can go in on the v4.1.x branch.

@wckzhang
Copy link
Contributor Author

These changes need to go in on master before they can go in on the v4.1.x branch.

The master branch changes haven't been merged either so I've been pushing fixes in to both PR's at the same time

@jsquyres jsquyres marked this pull request as draft July 28, 2020 17:47
@jsquyres
Copy link
Member

The master branch changes haven't been merged either so I've been pushing fixes in to both PR's at the same time

Cool. I just converted this PR to "draft" -- let's leave it as draft until the corresponding master PR has been merged and this PR is therefore ready for v4.1.x.

Thanks!

The default algorithm selections were out of date and not performing
well. After gathering data from OMPI developers, new default algorithm
decisions were selected for:

    allgather
    allgatherv
    allreduce
    alltoall
    alltoallv
    barrier
    bcast
    gather
    reduce
    reduce_scatter_block
    reduce_scatter
    scatter

These results were gathered using the ompi-collectives-tuning package
and then averaged amongst the results gathered from multiple OMPI
developers on their clusters.

You can access the graphs and averaged data here:
https://drive.google.com/drive/folders/1MV5E9gN-5tootoWoh62aoXmN0jiWiqh3

Signed-off-by: William Zhang <wilzhang@amazon.com>
(cherry picked from commit ce40cfb)
@wckzhang
Copy link
Contributor Author

The master branch changes haven't been merged either so I've been pushing fixes in to both PR's at the same time

Cool. I just converted this PR to "draft" -- let's leave it as draft until the corresponding master PR has been merged and this PR is therefore ready for v4.1.x.

Thanks!

Thanks Jeff, PR merged and I updated cherry pick ID.

@wckzhang wckzhang marked this pull request as ready for review July 28, 2020 22:49
@jsquyres
Copy link
Member

jsquyres commented Aug 3, 2020

Here's the corresponding master PR: #7730

@bwbarrett bwbarrett merged commit 704d019 into open-mpi:v4.1.x Aug 3, 2020
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Projects
None yet
Development

Successfully merging this pull request may close these issues.

5 participants