deepseek-ai/3FS
1
0
Fork 0
mirror of https://github.com/deepseek-ai/3FS.git synced 2026-11-05 00:47:47 +00:00
Code Issues 166 Projects Packages Wiki Activity

[PR #402] [Proposal] Design Doc of GDR integration for 3FS #416

New issue
Open
opened 2026-10-26 09:50:04 +00:00 by bsakaguchi · 0 comments
bsakaguchi commented 2026-10-26 09:50:04 +00:00
Owner
Copy link

📋 Pull Request Information

Original PR: https://github.com/deepseek-ai/3FS/pull/402
Author: @SimonCqk
Created: 3/17/2026
Status: 🔄 Open

Base: main ← Head: simoncqk/gdr-proposal-doc

📝 Commits (1)

  • e910e79 docs(proposal): add design proposal for GDR integrating into 3FS.

📊 Changes

1 file changed (+382 additions, -0 deletions)

View changed files

➕ docs/proposals/usrbio-gdr-integration-proposal.md (+382 -0)

📄 Description

As #395 discussed, currently, the application copies data to/from GPU VRAM. This adds latency, wastes CPU memory bandwidth, and can be a throughput bottleneck for model training and inference.

This proposal introduces GPU Direct RDMA (GDR) in 3FS, enabling the storage server to RDMA read/write directly to/from GPU VRAM with zero CPU-side copies.

I'd like to focus the discussion on the design first — collect community feedback on the API surface, architecture decisions, and known limitations before refining and submitting the complete implementation.

Waiting for your feedback and suggestions!

🔄 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/3FS/pull/402 **Author:** [@SimonCqk](https://github.com/SimonCqk) **Created:** 3/17/2026 **Status:** 🔄 Open **Base:** `main` ← **Head:** `simoncqk/gdr-proposal-doc` --- ### 📝 Commits (1) - [`e910e79`](https://github.com/deepseek-ai/3FS/commit/e910e794baf693be2a177e069b8b1cb93325ab58) docs(proposal): add design proposal for GDR integrating into 3FS. ### 📊 Changes **1 file changed** (+382 additions, -0 deletions) <details> <summary>View changed files</summary> ➕ `docs/proposals/usrbio-gdr-integration-proposal.md` (+382 -0) </details> ### 📄 Description As #395 discussed, currently, the application copies data to/from GPU VRAM. This adds latency, wastes CPU memory bandwidth, and can be a throughput bottleneck for model training and inference. This proposal introduces GPU Direct RDMA (GDR) in 3FS, enabling the storage server to RDMA read/write directly to/from GPU VRAM with zero CPU-side copies. I'd like to focus the discussion on the design first — collect community feedback on the API surface, architecture decisions, and known limitations before refining and submitting the complete implementation. Waiting for your feedback and suggestions! --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>
Sign in to join this conversation.
No Branch/Tag specified
Branches Tags
main
SF-Zhou-patch-1
No results found.
Labels
Clear labels   
pull-request
Mirrored from GitHub Pull Request

No labels
pull-request
Milestone
Clear milestone
No items
No milestone
Projects
Clear projects
No items
No project
Assignees
Clear assignees
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/3FS#416
No description provided.