SC2025Top-tier venue
Phoenix: A Refactored I/O Stack for GPU Direct Storage without Phony Buffers
Jianqin Yan, Shi Qiu, Yina Lv, Yifan Hu, Hao Chen, Zhirong Shen, Xin Yao, Renhai Chen, Jiwu Shu, Gong Zhang, Yiming Zhang
Abstract
GPU Direct Storage (GDS) plays a vital role in GPU-based training and inference systems, leveraging Peer-to-Peer Direct Memory Access (P2P-DMA) to establish a direct data transfer path between the GPU and the storage device. The direct I/O path reduces GPU storage access latency and CPU overhead, thus improving the efficiency of data transfer. Currently, however, GDS employs a phony buffer in the host memory to interact with the Linux kernel, which results in suboptimal I/O performance, extra resource consumption, and high deployment complexity.
Ask about this paper
Ask your agent about it.
Lune has read the top-tier papers around this one, so every answer names the papers it rests on.
Your agent calls
Lunesearch_papers
Free to start. No credit card required.
Terminal
Install the CLIlune papers get fe76a606-bca6-4bad-8dca-764745b44c21Cited by top-tier papers1
Ask how each one uses itRelated papers
- Managing Scalable Direct Storage Accesses for GPUs with GoFSShaobo Li, Yirui Eric Zhou, Yuqi Xue, Yuan Xu et al.SOSP 2025
- Efficient Multi-GPU Shared Memory via Automatic Optimization of Fine-Grained TransfersHarini Muthukrishnan, David W. Nellans, Daniel Lustig, Jeffrey A. Fessler et al.ISCA 2021 · 16 citations
- <u>G</u>PU <u>i</u>nitiated <u>O</u>penSHMEM: correct and efficient intra-kernel networking for dGPUsKhaled Hamidouche, Michael LeBeanePPoPP 2020 · 17 citations
- ARK: GPU-driven Code Execution for Distributed Deep LearningChangho Hwang, KyoungSoo Park, Ran Shu, Xinyuan Qu et al.NSDI 2023 · 22 citations
- RoPeerTo: A Datacenter-Scale Architecture for Peer-To-Peer DMA between GPUs and FPGAsMarco Venere, Giuseppe Sorrentino, Benjamin Ramhorst, Maximilian Jakob Heer et al.EuroSys 2026 · 1 citation
