<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>InfiniBand on Council of Elrond</title><link>https://songkou.github.io/tags/infiniband/</link><description>Recent content in InfiniBand on Council of Elrond</description><generator>Hugo</generator><language>en</language><lastBuildDate>Thu, 23 Jul 2026 21:00:00 +0800</lastBuildDate><atom:link href="https://songkou.github.io/tags/infiniband/index.xml" rel="self" type="application/rss+xml"/><item><title>RDMA Performance Tuning: From Fresh Install to Line-Rate Bandwidth</title><link>https://songkou.github.io/posts/rdma-performance-tuning/</link><pubDate>Thu, 23 Jul 2026 21:00:00 +0800</pubDate><guid>https://songkou.github.io/posts/rdma-performance-tuning/</guid><description>&lt;p&gt;The number-one complaint with RDMA is simple: the bandwidth won&amp;rsquo;t fill up. A 100G card pushing only 60G, two days spent checking drivers, optics, and the peer&amp;rsquo;s configuration — and in the end the fix is a single PCIe parameter.&lt;/p&gt;
&lt;p&gt;When RDMA underperforms, 90% of the time the cause is not exotic. It&amp;rsquo;s some factory default that was never touched: the switch buffer carved too small, the PCIe read request size left at its tiny default, the MTU misaligned on one segment of the path. None of these is hard to fix on its own — the hard part is knowing which one to check.&lt;/p&gt;</description></item></channel></rss>