Replace full EFA installer with minimal EFA + custom libfabric build #802
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
akkart-aws
requested review from
aranadive,
ovidiusm,
brminich and
a team
as code owners
|
👋 Hi akkart-aws! Thank you for contributing to ai-dynamo/nixl. Your PR reviewers will review your contribution then trigger the CI to test your changes. 🚀 |
|
/build |
|
/ok to test 904ab5a |
yosefe
previously approved these changes
Sep 17, 2025
ovidiusm
reviewed
Sep 18, 2025
ovidiusm
reviewed
Sep 18, 2025
ovidiusm
reviewed
Sep 18, 2025
akkart-aws
force-pushed
the
efa_installer_minimum
branch
2 times, most recently
from
a0433b3 to
c47556d
Compare
|
/ok to test c47556d |
|
/build |
Use compile_args instead of include_directories when libfabric_path is an absolute path to avoid meson build issues with external library includes.
|
/ok to test ed12136 |
|
/build |
Add hwloc and hwloc-devel packages to support hardware locality detection.
|
/ok to test 7889d3f |
|
/build |
|
/build |
|
/ok to test 6b13de5 |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
What?
Replace full EFA installer with minimal EFA installation and custom libfabric build from source. This change switches from using
efa_installer.sh -y -g(which installs OpenMPI, NCCL, and other unwanted packages) toefa_installer.sh -y --minimal(which installs only RDMA core and EFA drivers) combined with a custom libfabric v2.3.0 build configured specifically for EFA, CUDA, and GDRCopy support.Why?
The original EFA installer was installing unwanted components like OpenMPI and NCCL that were causing package conflicts and bloating the installation. The user specifically needed only EFA drivers and libfabric components without the additional packages that come with the full EFA installer.
How?
Modified EFA installer usage: Changed from
-y -gto-y --minimalflag to install only essential RDMA core and EFA driver componentsAdded custom libfabric build: Downloads libfabric v2.3.0 from GitHub releases and builds with specific configure options:
--enable-efafor EFA provider support--disable-verbs/psm3/opx/usnic/rstreamto disable unused providers--with-cuda=/usr/local/cuda --enable-cuda-dlopenfor CUDA support--with-gdrcopy --enable-gdrcopy-dlopenfor GDRCopy supportUpdated build system: Added
LIBFABRIC_VERSIONandLIBFABRIC_INSTALL_DIRparameters following the existing UCX pattern, updated environment variables and meson configuration to use custom libfabric pathsApplied consistently: Updated
.gitlab/build.sh,contrib/Dockerfile, andbenchmark/nixlbench/contrib/Dockerfilewith the same approach