<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:content="http://purl.org/rss/1.0/modules/content/">
  <channel>
    <title>P2P-DMA on Hellokitty&#39;s Blog</title>
    <link>https://yangyang233333.github.io/tags/p2p-dma/</link>
    <description>Recent content in P2P-DMA on Hellokitty&#39;s Blog</description>
    <generator>Hugo</generator>
    <language>zh-CN</language>
    <lastBuildDate>Mon, 24 Aug 2026 14:49:50 +0800</lastBuildDate>
    <atom:link href="https://yangyang233333.github.io/tags/p2p-dma/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>Phoenix 论文阅读：一种不使用伪缓冲区的 GPU Direct Storage 重构方案</title>
      <link>https://yangyang233333.github.io/posts/phoenix-paper-chinese-translation/</link>
      <pubDate>Mon, 24 Aug 2026 14:49:50 +0800</pubDate>
      <guid>https://yangyang233333.github.io/posts/phoenix-paper-chinese-translation/</guid>
      <description>按原论文结构详译 SC&#39;25 论文 Phoenix：解释 NVIDIA GDS 的伪缓冲区问题、基于 ZONE_DEVICE 的 GPU 显存映射、POSIX/io_uring 编程模型，以及本地 NVMe、远程存储、KV Cache 和模型加载实验。</description>
    </item>
  </channel>
</rss>
