Performance Tuning & Kernel Parameters (Sysctl)

Stop letting your high-traffic web server drop packets with listen backlog tweaks

🧩 The Challenge

You ever see your web server logs reporting connection refused errors even though the app isn’t even touching the CPU limits? I spent half a morning once convinced the app was crashing, only to realize the OS was just throwing away connections because the queue for pending requests was full.

💡 The Fix

You need to bump up the backlog queue depth so the kernel has a bigger waiting room for incoming connections before they get handed off to your application. This is a life saver during sudden traffic bursts that would otherwise cause dropped connections.

sysctl -w net.core.somaxconn=4096
sysctl -w net.ipv4.tcp_max_syn_backlog=4096

⚙️ Why It Works

Most kernels ship with a default backlog of 128, which is basically an empty parking space for a modern web server. By bumping these values, you’re letting the TCP stack hold onto those pending handshakes long enough for your worker threads to actually wake up and process them.

🚀 Pro-Tip: Don’t forget to also update your web server’s internal config file to match, otherwise, the kernel will just clip the request anyway.

Linux Tips & Tricks | © ngelinux.com | 8/25/2026