perf(combine): schedule scale-out puts remote-first to mitigate node variance - #8
Open
Xuan-1998 wants to merge 1 commit into
Open
Conversation
…stagger The combine forward warps replayed tokens in the dispatch arrival order frozen into the handle, which interleaves local-bypass tokens (no RDMA) with remote tokens in a network-timing-dependent way. A node whose replay front-loads local tokens back-loads all of its scale-out puts so with the NIC near saturation the shift never drains and surfaces as the peer node's exit-wait tail, so one node runs ~62 GB/s SO while the other runs ~76. Here we make the schedule deterministic: the scale-up warps sweep the linked list twice (remote-destined tokens first, local-bypass second) and the forward warps replay in the same two-pass order, so count-based tail gating is unchanged and every rank issues its puts on the same schedule. Pass 0 flushes its TMA record and remainder batches before pass 1 so the wire keeps draining while locals are reduced.
Xuan-1998
changed the title
|
This commit also exist in https://github.com/amazon-contributing/DeepEP/pull/9/commits. |
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
The combine forward warps replayed tokens in the dispatch arrival order frozen into the handle, which interleaves local-bypass tokens (no RDMA) with remote tokens in a network timing dependent way. A node whose replay front loads local tokens back loads all of its scale-out puts will cause its peer node's exit-wait tail, so one node runs ~62 GB/s SO (waiting for the remote tokens to arrive since the local tokens were processed first) while the other runs ~76 GB/s SO.
So here we made both scaleup and forward warps processing remote tokens first and then local tokens to mitigate the issue.