Boost Nginx Concurrency: Master the 4 Key Performance Parameters

Worker Processes and Connections

The worker_processes directive sets the number of Nginx worker processes. It is usually matched to the number of CPU cores or set to auto so Nginx detects the optimal count automatically. worker_processes auto; The worker_connections directive defines how many simultaneous connections each worker can handle. The product of worker_processes and worker_connections approximates the maximum concurrent connections Nginx can support under ideal conditions. worker_connections 65535; Actual limits are also constrained by operating‑system parameters such as file‑descriptor limits (ulimit) and epoll settings, so OS‑level tuning is required for the theoretical maximum to be reachable.

Keep‑Alive Settings

The keepalive_timeout directive controls how long an idle client connection is kept open. A reasonable timeout reuses TCP connections for repeated requests, reducing handshake overhead and improving throughput. keepalive_timeout 65; The keepalive_requests directive sets the maximum number of requests that can be sent over a single keep‑alive connection. The default is 100; increasing it (e.g., to 10000) can further reduce connection churn. keepalive_requests 10000; Timeout values must balance resource consumption against the performance gains from connection reuse; overly long timeouts waste connections, while too short timeouts diminish reuse benefits.

High‑Performance I/O Options

Three common I/O optimizations significantly improve static‑resource delivery and latency: sendfile enables zero‑copy transmission in the kernel, eliminating user‑space data copies. tcp_nopush (or TCP_CORK) merges small packets when sending large files, increasing bandwidth utilization. tcp_nodelay disables Nagle’s algorithm, which is beneficial for low‑latency, small‑packet interactions.

Combining these settings with appropriate worker_processes, worker_connections, and keep‑alive parameters allows Nginx to achieve higher throughput and lower latency across diverse workloads.