ofi: Share the domain among MTL and BTL #13415
Open
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
shijin-aws
reviewed
Sep 24, 2025
shijin-aws
approved these changes
Sep 24, 2025
|
@bwbarrett @hppritcha Can you review this? |
bwbarrett
requested changes
Sep 29, 2025
jiaxiyan
changed the title
bwbarrett
requested changes
Oct 8, 2025
bwbarrett
reviewed
Oct 8, 2025
jiaxiyan
changed the title
bwbarrett
previously approved these changes
Oct 9, 2025
bwbarrett
requested changes
Oct 15, 2025
There was a problem hiding this comment.
Please reorder the commits so that the domain sharing patch is last. The others are cleaning up bugs that are more likely to be exposed when the sharing code runs (as we've found with our efa provider cleanups).
For the share the domain patch, in the commit message, please add some details on why we did this as opposed to just having the common code allocate a domain with the superset of requirements. Namely, that in this case, if providers can reuse the fabric and domain given the different capability sets, we'll do that. But if providers can't reuse the fabric and domain, we'll happily just create two domains.
Other than that, looks god to me.
Signed-off-by: Jessie Yang <jiaxiyan@amazon.com>
Add FI_COMPLETION flag to ensure completion entries are generated for all data transfer operations. Signed-off-by: Jessie Yang <jiaxiyan@amazon.com>
Share the domain between the MTL and BTL layers to reduce the total number of domains created. This helps avoid hitting system resource limits on platforms with high core counts. Instead of having the common code allocate a single domain with the superset of all required capabilities, we attempt to reuse an existing fabric and domain if the providers can support MTL’s and BTL’s different capability sets. This approach allows providers that support domain sharing to reuse resources efficiently while still preserving flexibility. If the providers cannot reuse the fabric and domain due to incompatible requirements, separate domains will be created as before. Signed-off-by: Jessie Yang <jiaxiyan@amazon.com>
shijin-aws
approved these changes
Oct 15, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
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.
Share the domain between the MTL and BTL layers to reduce the total number of domains created. This helps avoid hitting system resource limits on platforms with high core counts.
Instead of having the common code allocate a single domain with the superset of all required capabilities, we attempt to reuse an existing fabric and domain if the providers can support MTL’s and BTL’s different capability sets. This approach allows providers that support domain sharing to reuse resources efficiently while still preserving flexibility. If the providers cannot reuse the fabric and domain due to incompatible requirements, separate domains will be created as before.