Skip to content
This repository was archived by the owner on Sep 30, 2022. It is now read-only.

Conversation

@hjelmn
Copy link
Member

@hjelmn hjelmn commented May 26, 2016

This commit reduces the default exclusivity so that btl/scif is not
used for send/recv over other shared memory transports.

Fixes open-mpi/ompi#1712

Signed-off-by: Nathan Hjelm hjelmn@lanl.gov

(cherry picked from open-mpi/ompi@5caf12c)

Signed-off-by: Nathan Hjelm hjelmn@lanl.gov

This commit reduces the default exclusivity so that btl/scif is not
used for send/recv over other shared memory transports.

Fixes open-mpi/ompi#1712

Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>

(cherry picked from open-mpi/ompi@5caf12c)

Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
@hjelmn
Copy link
Member Author

hjelmn commented May 26, 2016

:bot🏷️bug
:bot:milestone:v2.0.0
:bot:assign: @abouteiller

@hppritcha Very minor bug affecting out-of-the-box performance when scif is installed. btl/scif was never intended to beat vader or sm.

@mellanox-github
Copy link

Test PASSed.
See http://bgate.mellanox.com/jenkins/job/gh-ompi-release-pr/1721/ for details.

@hppritcha hppritcha modified the milestones: v2.0.1, v2.0.0 May 27, 2016
@abouteiller
Copy link
Member

@hjelmn

The problem is only partially fixed: because in 2.x multirail is on by default, even when SCIF does not have a high exclusivity, it gets loaded for larger message sizes where the message is stripped on multiple transports (1 frag on vader, 1 frag on scif), leading again to poor performance in that setup.

@hjelmn
Copy link
Member Author

hjelmn commented May 27, 2016

@abouteiller Can you try setting btl_scif_latency to a higher number and see if that helps. 100000 should be high enough.

@abouteiller
Copy link
Member

the remaining problem is in TCP btl issue, so this is a correct fix without further work needed.
👍

@abouteiller
Copy link
Member

@hppritcha This is minor change, it has already been integrated in 1.10, I think it should be integrated in v2.0.0.

@jsquyres
Copy link
Member

Discussed on 2016-06-14 webex; approved for v2.0.0.

@jsquyres jsquyres modified the milestones: v2.0.0, v2.0.1 Jun 14, 2016
@jsquyres jsquyres merged commit 8b0af6a into open-mpi:v2.x Jun 14, 2016
Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.

Projects

None yet

Development

Successfully merging this pull request may close these issues.

6 participants