How to Tune Nginx Kernel Parameters for Up to 10× Performance Boost

Process and Connection Optimization

Nginx’s worker_processes is set to auto to match CPU cores, and worker_cpu_affinity is also auto to bind each worker to a specific core, reducing context switches. The worker_connections directive is raised to 65535, defining the maximum concurrent connections per worker. The event model is explicitly set to epoll and multi_accept on to accept as many new connections as possible after a notification.

# nginx.conf
worker_processes  auto;
worker_cpu_affinity  auto;
events {
    worker_connections 65535;  # max connections per worker
    use epoll;               # use epoll event model
    multi_accept on;         # accept many connections at once
}

System‑wide Connection Queue Optimization

Because Nginx cannot exceed the operating‑system limits, the article updates /etc/security/limits.conf to raise both soft and hard nofile values to 655350. It also modifies /etc/sysctl.conf to increase the global listen queue ( net.core.somaxconn=65535) and the SYN backlog ( net.ipv4.tcp_max_syn_backlog=65535). Finally, each server block’s listen directive specifies backlog=65535 instead of the default 511.

# /etc/security/limits.conf
* soft nofile 655350
* hard nofile 655350

# /etc/sysctl.conf
net.core.somaxconn = 65535
net.ipv4.tcp_max_syn_backlog = 65535

# nginx.conf (inside server block)
listen 80 backlog=65535;

Zero‑Copy and Buffer Optimizations

Enabling sendfile on allows the kernel to transfer files directly from disk to the network interface, bypassing user‑space buffers and dramatically reducing CPU usage. The article also turns on tcp_nopush on to coalesce data into full TCP packets before sending, and tcp_nodelay on to transmit partially‑filled packets immediately, improving latency. The keep‑alive timeout is increased to 65 seconds.

# http {
    sendfile on;          # zero‑copy file transfer
    tcp_nopush on;        # pack data into full TCP segments
    tcp_nodelay on;       # send small packets without delay
    keepalive_timeout 65;
}

Combined Effect

By aligning Nginx’s internal limits with the operating system and leveraging zero‑copy plus TCP optimizations, the configuration can raise throughput by an order of magnitude while keeping CPU utilization low. The article demonstrates the concrete configuration changes required to achieve this performance gain.