[PR #875] [MERGED] feat(store): Add multi threading handle page fault during segment allocation #1880

New issue

Closed

opened 2026-10-29 21:31:04 +00:00 by bsakaguchi · 0 comments

bsakaguchi commented

2026-10-29 21:31:04 +00:00

Owner

📋 Pull Request Information

Original PR: https://github.com/kvcache-ai/Mooncake/pull/875
Author: @xiaguan
Created: 9/22/2025
Status: ✅ Merged
Merged: 9/28/2025
Merged by: @ykwd

Base: main ← Head: speedup_segment

📝 Commits (10+)

e328993 feat(store): Add multi threading during segment allocation
618540c remove samll pre touch
0d97623 Temp version
e7db9dd temp version
49e142f temp version
8c9837a temp version
18974d1 Update log
4523e5f new version
e505cd8 Remove timer log
a4e24f9 Format code

📊 Changes

7 files changed (+141 additions, -14 deletions)

View changed files

📝 mooncake-transfer-engine/include/memory_location.h (+5 -1)
📝 mooncake-transfer-engine/include/transport/rdma_transport/rdma_context.h (+6 -0)
📝 mooncake-transfer-engine/include/transport/rdma_transport/rdma_transport.h (+2 -0)
📝 mooncake-transfer-engine/src/memory_location.cpp (+6 -2)
📝 mooncake-transfer-engine/src/transport/rdma_transport/rdma_context.cpp (+21 -3)
📝 mooncake-transfer-engine/src/transport/rdma_transport/rdma_transport.cpp (+91 -4)
📝 mooncake-transfer-engine/tests/memory_location_test.cpp (+10 -4)

📄 Description

bench mount 180GB segment
main: 74s
this pr: 46s(31s for register RDMA)

A 10GB threshold is necessary. I've tried with 1GB, and having too many threads actually slows things down. 10GB works pretty well in testing.

This will greatly improve the experience for users mounting TB-sized segments, helping them start up much faster.

related issue #848

_{🔄 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/kvcache-ai/Mooncake/pull/875 **Author:** [@xiaguan](https://github.com/xiaguan) **Created:** 9/22/2025 **Status:** ✅ Merged **Merged:** 9/28/2025 **Merged by:** [@ykwd](https://github.com/ykwd) **Base:** `main` ← **Head:** `speedup_segment` --- ### 📝 Commits (10+) - [`e328993`](https://github.com/kvcache-ai/Mooncake/commit/e328993b0f19aca7f90e04ef4eacc8bd35df569e) feat(store): Add multi threading during segment allocation - [`618540c`](https://github.com/kvcache-ai/Mooncake/commit/618540cc20271f43d64af82ae7676b0583d4b0e1) remove samll pre touch - [`0d97623`](https://github.com/kvcache-ai/Mooncake/commit/0d97623fd691e189103982a4801f9706191f0139) Temp version - [`e7db9dd`](https://github.com/kvcache-ai/Mooncake/commit/e7db9dd85a2e93cecf3baa3d6a3c132eab76929f) temp version - [`49e142f`](https://github.com/kvcache-ai/Mooncake/commit/49e142fa1dca37d00cbc4af50f2b4124cdc005bb) temp version - [`8c9837a`](https://github.com/kvcache-ai/Mooncake/commit/8c9837a78acb86569e411d2c66b2aafe34e94134) temp version - [`18974d1`](https://github.com/kvcache-ai/Mooncake/commit/18974d1bc0cd3edbb6fd11b9e44ed1d503560634) Update log - [`4523e5f`](https://github.com/kvcache-ai/Mooncake/commit/4523e5f04eaf87ae7e4c56de5411f7d119c77e24) new version - [`e505cd8`](https://github.com/kvcache-ai/Mooncake/commit/e505cd8bcfefdf99f42a3d153ff527bf6d9a65d6) Remove timer log - [`a4e24f9`](https://github.com/kvcache-ai/Mooncake/commit/a4e24f99ce091f67fac9b0916c4bfcf75aa69e8f) Format code ### 📊 Changes **7 files changed** (+141 additions, -14 deletions) <details> <summary>View changed files</summary> 📝 `mooncake-transfer-engine/include/memory_location.h` (+5 -1) 📝 `mooncake-transfer-engine/include/transport/rdma_transport/rdma_context.h` (+6 -0) 📝 `mooncake-transfer-engine/include/transport/rdma_transport/rdma_transport.h` (+2 -0) 📝 `mooncake-transfer-engine/src/memory_location.cpp` (+6 -2) 📝 `mooncake-transfer-engine/src/transport/rdma_transport/rdma_context.cpp` (+21 -3) 📝 `mooncake-transfer-engine/src/transport/rdma_transport/rdma_transport.cpp` (+91 -4) 📝 `mooncake-transfer-engine/tests/memory_location_test.cpp` (+10 -4) </details> ### 📄 Description bench mount 180GB segment main: 74s this pr: 46s(31s for register RDMA) A 10GB threshold is necessary. I've tried with 1GB, and having too many threads actually slows things down. 10GB works pretty well in testing. This will greatly improve the experience for users mounting TB-sized segments, helping them start up much faster. related issue #848 --- <sub>🔄 This issue represents a GitHub Pull Request. It cannot be merged through Gitea due to API limitations.</sub>

bsakaguchi

2026-10-29 21:31:04 +00:00