new commit

networking/rss-irq-affinity-and-rps/index.html

@@ -3245,8 +3245,8 @@ <h3 id="using-flamegraphs-to-analyze-cpu-activity">Using flamegraphs to analyze
 the server should be running on CPUs 6 and 7.
 There are 2 reasons for this:</p>
 <ul>
-<li>We started the application on CPUs 6 and 7 which are isolated from via <code class="highlight">isolcpus</code> in <code class="highlight">/proc/cmdline</code>, so the kernel
-loadbalancer is off for these CPUs</li>
+<li>We started the application on CPUs 6 and 7 which are isolated via <code class="highlight">isolcpus</code> in <code class="highlight">/proc/cmdline</code>, so the kernel
+loadbalancer is off for these CPUs.</li>
 <li>Have another look at the server implementation: you will see that the TCP server is multithreaded (use of go routines)
 and the UDP server is single threaded. I had not intended it to be this way - it was an omission on my side because
 I initially implemented the TCP code and then quickly added the UDP part. This is fixed in the master branch of the
@@ -3282,7 +3282,7 @@ <h3 id="using-flamegraphs-to-analyze-cpu-activity">Using flamegraphs to analyze
 <span class="p">(</span><span class="o">...</span><span class="p">)</span>
 </code></pre></div></td></tr></table></div>
 <p><img alt="IRQ smp affinity - flamegraph 6" src="https://github.com/user-attachments/assets/8d796b41-b056-4048-b5d4-5385538ecdb2" /></p>
-<p>CPUs 2 and 3 process our softirqs, at roughly 20% and 13% respectively. Actually, running top or mpstat during the
+<p>CPUs 2 and 3 process our softirqs, at roughly 20% and 13% respectively. Actually, running <code class="highlight">top</code> or <code class="highlight">mpstat</code> during the
 tests also showed that the CPUs were spending that amount of time processing hardware interrupts and softirqs.</p>
 <p>This is pure speculation, but I suppose that we do not see any hardware interrupts here for 2 reasons:</p>
 <ul>