Boost Nginx Concurrency: Master the 4 Core Parameters for 100k+ Connections

1. worker_processes (Number of worker processes)

This setting determines how many CPU cores Nginx can utilize. Recommended values are the number of physical cores or auto (supported from Nginx 1.11+). Example:

<ol><li><code># Recommended: equal to CPU cores or auto</code></li><li><code>worker_processes auto;</code></li><li><code># Or set manually (recommended for production)</code></li><li><code>worker_processes 8;  # adjust to your CPU cores</code></li></ol>

Best practice: worker_processes = CPU physical cores × (1~2), using 2× for hyper‑threaded CPUs.

2. worker_connections (Maximum connections per worker)

Defines the maximum concurrent connections each worker can handle.

<ol><li><code>events {</code></li><li><code># default is 1024 – too low for high traffic</code></li><li><code>    worker_connections 10240;  # common range: 10240‑65535</code></li><li><code>    # or higher if system limits allow</code></li><li><code>    worker_connections 32768;</code></li><li><code>}</code></li></ol>

The theoretical maximum concurrency equals worker_processes × worker_connections. For example, 8 processes × 10240 connections = 81,920 concurrent connections.

3. System‑level connection limits (ulimit and sysctl)

Nginx is also constrained by the operating system’s file‑descriptor and network limits.

Modify user‑level limits (ulimit):

<ol><li><code># Permanently edit /etc/security/limits.conf</code></li><li><code>* soft nofile 1048576</code></li><li><code>* hard nofile 1048576</code></li><li><code>root soft nofile 1048576</code></li><li><code>root hard nofile 1048576</code></li></ol>

Modify kernel parameters (sysctl):

<ol><li><code># Add to /etc/sysctl.conf</code></li><li><code>fs.file-max = 1048576</code></li><li><code>net.core.somaxconn = 32768</code></li><li><code>net.ipv4.tcp_max_syn_backlog = 16384</code></li><li><code>net.core.netdev_max_backlog = 16384</code></li><li><code># TCP reuse optimizations (recommended)</code></li><li><code>net.ipv4.tcp_tw_reuse = 1</code></li><li><code>net.ipv4.tcp_tw_recycle = 0  # deprecated, do not enable</code></li><li><code>net.ipv4.tcp_fin_timeout = 15</code></li><li><code>net.ipv4.tcp_keepalive_time = 300</code></li></ol>

4. multi_accept (Accept multiple connections at once)

Enables each worker to accept many new connections in a single batch instead of one‑by‑one.

<ol><li><code>events {</code></li><li><code>    use epoll;  # Linux‑recommended</code></li><li><code>    worker_connections 10240;</code></li><li><code>    multi_accept on;  # default off – must enable</code></li><li><code>}</code></li></ol>

Final high‑concurrency configuration (typical for 100k+ connections)

worker_processes auto;  # or CPU cores ×2
worker_rlimit_nofile 1048576;  # override ulimit

events {
    use epoll;
    worker_connections 32768;
    multi_accept on;
}

http {
    sendfile on;
    tcp_nopush on;
    tcp_nodelay on;
    keepalive_timeout 65;
    keepalive_requests 100000;
    limit_conn_zone $binary_remote_addr zone=addr:10m;
    limit_conn addr 50;
    # additional optimizations can be added here
}

To verify the effective limits, you can check Nginx’s maximum open files and current connections:

# Show Nginx max open files
cat /proc/$(cat /run/nginx.pid)/limits | grep "Max open files"

# Show current system connections
ss -s
netstat -n | awk '/^tcp/ {++S[$NF]} END {for(a in S) print a, S[a]}'