<?xml version="1.0" encoding="utf-8" standalone="yes"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom"><channel><title>Performance on Le Site de François</title><link>https://lesitedefrancois.be/en/tags/performance/</link><description>Recent content in Performance on Le Site de François</description><generator>Hugo -- gohugo.io</generator><language>en</language><copyright>© 2026 François</copyright><atom:link href="https://lesitedefrancois.be/en/tags/performance/index.xml" rel="self" type="application/rss+xml"/><item><title>DPDK</title><link>https://lesitedefrancois.be/en/telco/dpdk/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://lesitedefrancois.be/en/telco/dpdk/</guid><description>&lt;p&gt;The &lt;strong&gt;Data Plane Development Kit (DPDK)&lt;/strong&gt; is an &lt;strong&gt;open-source&lt;/strong&gt; set of libraries and &lt;strong&gt;poll-mode drivers (PMDs)&lt;/strong&gt; that move &lt;strong&gt;packet processing&lt;/strong&gt; from the kernel to &lt;strong&gt;userspace&lt;/strong&gt;, enabling telco and cloud applications to achieve &lt;strong&gt;millions of packets per second&lt;/strong&gt; per core with &lt;strong&gt;predictable latency&lt;/strong&gt;. DPDK bypasses the traditional socket stack: applications &lt;strong&gt;busy-poll&lt;/strong&gt; NIC queues (or &lt;strong&gt;virtio/vhost&lt;/strong&gt; rings), use &lt;strong&gt;hugepages&lt;/strong&gt; to reduce TLB misses, and pin threads to &lt;strong&gt;NUMA-local&lt;/strong&gt; cores — a model suited to &lt;strong&gt;UPF&lt;/strong&gt;, &lt;strong&gt;vRouter&lt;/strong&gt;, &lt;strong&gt;CG-NAT&lt;/strong&gt;, &lt;strong&gt;load balancers&lt;/strong&gt;, and &lt;strong&gt;5G user-plane&lt;/strong&gt; functions where per-packet syscall overhead is unacceptable.&lt;/p&gt;</description></item><item><title>SR-IOV</title><link>https://lesitedefrancois.be/en/telco/sr-iov/</link><pubDate>Mon, 01 Jan 0001 00:00:00 +0000</pubDate><guid>https://lesitedefrancois.be/en/telco/sr-iov/</guid><description>&lt;p&gt;&lt;strong&gt;SR-IOV (Single Root I/O Virtualisation)&lt;/strong&gt; is a &lt;strong&gt;PCI-SIG&lt;/strong&gt; specification that lets one physical &lt;strong&gt;PCIe&lt;/strong&gt; device (typically a &lt;strong&gt;NIC&lt;/strong&gt; or accelerator) expose multiple lightweight &lt;strong&gt;Virtual Functions (VFs)&lt;/strong&gt; — each assignable directly to a &lt;strong&gt;VM&lt;/strong&gt; or container — while a &lt;strong&gt;Physical Function (PF)&lt;/strong&gt; remains for management and global configuration. VFs bypass much of the hypervisor’s software switching path, delivering &lt;strong&gt;lower latency&lt;/strong&gt;, &lt;strong&gt;higher throughput&lt;/strong&gt;, and more &lt;strong&gt;deterministic&lt;/strong&gt; behaviour than &lt;strong&gt;paravirtualised virtio&lt;/strong&gt; alone — properties valued in &lt;strong&gt;telco NFV&lt;/strong&gt; (vEPC, vRAN CU/DU, firewall, DPI) and in &lt;strong&gt;cloud-native&lt;/strong&gt; packet workloads on &lt;strong&gt;Kubernetes&lt;/strong&gt;.&lt;/p&gt;</description></item></channel></rss>