Boost Nginx Concurrency: Key Settings and Practical Configurations

worker_processes – Upper limit of concurrency

worker_processes defines how many worker processes Nginx can spawn; setting it to auto makes Nginx create one worker per CPU core, maximizing CPU utilization while avoiding excessive context switches.

worker_processes auto;

worker_connections – Connections per worker

worker_connections specifies the maximum number of simultaneous connections a single worker can handle. The theoretical maximum concurrent connections is calculated as:

max_concurrency = worker_processes * worker_connections;

For example, 8 workers × 10240 connections = 81 920 concurrent connections. A typical recommended setting is: worker_connections 10240; Note: the operating system’s ulimit -n must be increased accordingly, otherwise the setting will not take effect.

keepalive_timeout – Tuning persistent connections

Long‑lived keep‑alive connections occupy slots, reducing overall concurrency and QPS. Reducing keepalive_timeout frees connections for new clients. A common value is 15 seconds, but for short‑lived traffic (e‑commerce APIs) it can be lowered to 3–10 seconds.

keepalive_timeout 15;

High‑Performance I/O Trio

These directives are especially useful for serving static assets (images, videos, etc.) at high throughput.

sendfile on;
tcp_nopush on;
tcp_nodelay on;

sendfile enables zero‑copy file transmission, reducing CPU load. tcp_nopush merges small packets to improve throughput. tcp_nodelay disables Nagle’s algorithm, lowering latency for fast‑response services. Combined, they provide high throughput with low latency.