[PR #594] [CLOSED] ll dispatch: bulk put per expert to reduce RDMA overhead #649

Closed
opened 2026-10-27 21:04:52 +00:00 by bsakaguchi · 0 comments
Owner

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/DeepEP/pull/594
Author: @Xuan-1998
Created: 3/23/2026
Status: ❌ Closed

Base: main ← Head: bulk-put-pr


📝 Commits (1)

  • 5029266 ll dispatch: bulk put per expert to reduce RDMA overhead

📊 Changes

1 file changed (+43 additions, -18 deletions)

View changed files

📝 csrc/kernels/internode_ll.cu (+43 -18)

📄 Description

Stage dispatch tokens into per-expert contiguous regions in the send buffer, then issue one bulk RDMA put per remote expert instead of one per token. This reduces the number of RDMA operations and proxy channel round-trips for the low-latency dispatch path.


🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.

## 📋 Pull Request Information **Original PR:** https://github.com/deepseek-ai/DeepEP/pull/594 **Author:** [@Xuan-1998](https://github.com/Xuan-1998) **Created:** 3/23/2026 **Status:** ❌ Closed **Base:** `main` ← **Head:** `bulk-put-pr` --- ### 📝 Commits (1) - [`5029266`](https://github.com/deepseek-ai/DeepEP/commit/5029266200a8c37505b33877c57138d75d3dd081) ll dispatch: bulk put per expert to reduce RDMA overhead ### 📊 Changes **1 file changed** (+43 additions, -18 deletions) <details> <summary>View changed files</summary> 📝 `csrc/kernels/internode_ll.cu` (+43 -18) </details> ### 📄 Description Stage dispatch tokens into per-expert contiguous regions in the send buffer, then issue one bulk RDMA put per remote expert instead of one per token. This reduces the number of RDMA operations and proxy channel round-trips for the low-latency dispatch path. --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>
bsakaguchi 2026-10-27 21:04:52 +00:00
Sign in to join this conversation.
No labels
pull-request
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
deepseek-ai/DeepEP#649
No description provided.