Performance Tuning

Workers and Connections

worker_processes auto;          # one per CPU core
worker_rlimit_nofile 65535;     # raise fd limit for high concurrency

events {
    worker_connections 4096;
    use epoll;
    multi_accept on;
}

Rough ceiling: worker_processes x worker_connections simultaneous connections.

Keepalive to Backends

Reuse backend connections instead of opening a new one per request:

upstream app_servers {
    server 127.0.0.1:8080;
    keepalive 32;
}
location / {
    proxy_http_version 1.1;
    proxy_set_header Connection "";
    proxy_pass http://app_servers;
}

Client-Side Keepalive

keepalive_timeout 65;
keepalive_requests 1000;   # many requests per single connection

Static-Only Optimizations

• sendfile on; (default) - kernel zero-copy file serving.
• tcp_nopush on; - pack responses into fewer packets.
• tcp_nodelay on; - low latency for interactive traffic.

Measure, Don't Guess

# traffic and status over time
sudo apt install -y nginx-module-sts nginx-extras  # or parse the access log
ab -n 10000 -c 100 http://localhost/   # quick load test 

Watch error.log for worker_connections limit warnings and RAM/swap for worker footprint.

Key Points

• Tune workers and file descriptors upward for concurrency.
• keepalive to backends removes a huge TCP overhead.
• Base every knob on measured traffic, not rumor.