<?xml version="1.0" encoding="utf-8" standalone="yes"?>
<rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom">
  <channel>
    <title>Dcqcn on Flavio Castro</title>
    <link>https://castroflavio.com/tags/dcqcn/</link>
    <description>Recent content in Dcqcn on Flavio Castro</description>
    <generator>Hugo</generator>
    <language>en</language>
    <copyright>&lt;a href=&#34;https://creativecommons.org/licenses/by-nc/4.0/&#34; target=&#34;_blank&#34; rel=&#34;noopener&#34;&gt;CC BY-NC 4.0&lt;/a&gt;</copyright>
    <lastBuildDate>Thu, 12 Nov 2026 00:00:00 +0000</lastBuildDate>
    <atom:link href="https://castroflavio.com/tags/dcqcn/index.xml" rel="self" type="application/rss+xml" />
    <item>
      <title>WTH is a lossless network?</title>
      <link>https://castroflavio.com/posts/wth-is-a-lossless-network/</link>
      <pubDate>Thu, 12 Nov 2026 00:00:00 +0000</pubDate>
      <guid>https://castroflavio.com/posts/wth-is-a-lossless-network/</guid>
      <description>&lt;p&gt;In this article I explore a few specific behaviors of modern AI networks, building on the work of two papers: &lt;em&gt;Congestion Control for Large-Scale RDMA Deployments&lt;/em&gt; (Zhu et al., SIGCOMM 2015), which introduced DCQCN and described the victim-flow problem, and &lt;em&gt;HPCC: High Precision Congestion Control&lt;/em&gt; (Li et al., SIGCOMM 2019), whose simulator we use.&lt;/p&gt;&#xA;&lt;h2 id=&#34;1-background&#34;&gt;1. Background&lt;/h2&gt;&#xA;&lt;p&gt;Recently, I was intrigued by incast storms and how they can cause micro-bursts of congestion. I thought this was normal TCP behavior and expected TCP fairness to do its magic. This led me to look at how ECN could improve this scenario, which I thought was the primary tool used in modern AI fabrics. But then I realized lots of these AI fabrics don&amp;rsquo;t even use TCP. That sent me to find out what all the fuss about RoCEv2 is.&lt;/p&gt;</description>
    </item>
  </channel>
</rss>
