caddy vs nginx: I Benchmarked Both Serving Static Files

A client asked me last week whether Caddy can handle production traffic the same way nginx does. I had opinions but no numbers. So I installed both on the same VM, served the same static files, and ran ApacheBench against each. Here is what I found.

nginx: nginx.org · GitHub
Caddy: caddyserver.com · GitHub

What You Need

1. A Debian/Ubuntu server (I used Ubuntu 26.04 on a 2 vCPU VM)
2. sudo access
3. ApacheBench (apt install apache2-utils) for the benchmark runs

Setup

I created a synthetic 1.5 KB static page and a 10 KB binary file, wrote minimal configs for both servers, and benchmarked with ab -n 5000 -c 100 on loopback. The test data is synthetic and uses RFC 5737-safe addresses for reproducibility.

Nginx Config

worker_processes 1;
events { worker_connections 1024; }
http {
    access_log off;
    server {
        listen 8091;
        root /var/www/bench;
    }
}

worker_processes 1;
events { worker_connections 1024; }
http {
    access_log off;
    server {
        listen 8091;
        root /var/www/bench;
    }
}

Caddy Config

8092 {
    root * /var/www/bench
    file_server
}

8092 {
    root * /var/www/bench
    file_server
}

Reverse Proxy Config Comparison

Both tools are also commonly used as reverse proxies. Here is the difference in config lines:

# nginx — 5 lines
server {
    listen 127.0.0.1:3000;
    location / {
        proxy_pass http://127.0.0.1:8080;
    }
}

# nginx — 5 lines
server {
    listen 127.0.0.1:3000;
    location / {
        proxy_pass http://127.0.0.1:8080;
    }
}

# Caddy — 1 line
127.0.0.1:3000 { reverse_proxy 127.0.0.1:8080 }

# Caddy — 1 line
127.0.0.1:3000 { reverse_proxy 127.0.0.1:8080 }

Benchmark Results

I ran ab -n 5000 -c 100 against each server. The full outputs:

# nginx 1.28.3
Requests per second:  15850.02 [#/sec] (mean)
Complete requests:      5000
Failed requests:           0
Transfer rate:        26839.78 [Kbytes/sec] received

Connection Times (ms)
              min   avg   max
Connect:        0     2    4
Processing:     1     4    7
Total:          4     6    8

Percentage of the requests served within a certain time (ms)
  50%      6
  95%      7
  99%      8

# nginx 1.28.3
Requests per second:  15850.02 [#/sec] (mean)
Complete requests:      5000
Failed requests:           0
Transfer rate:        26839.78 [Kbytes/sec] received

Connection Times (ms)
              min   avg   max
Connect:        0     2    4
Processing:     1     4    7
Total:          4     6    8

Percentage of the requests served within a certain time (ms)
  50%      6
  95%      7
  99%      8

# Caddy v2.11.9
Requests per second:   7524.22 [#/sec] (mean)
Complete requests:      5000
Failed requests:           0
Transfer rate:        12858.77 [Kbytes/sec] received

Connection Times (ms)
              min   avg   max
Connect:        0     1   11
Processing:     0    13   48
Total:          0    13   48

Percentage of the requests served within a certain time (ms)
  50%     12
  95%     23
  99%     29

# Caddy v2.11.9
Requests per second:   7524.22 [#/sec] (mean)
Complete requests:      5000
Failed requests:           0
Transfer rate:        12858.77 [Kbytes/sec] received

Connection Times (ms)
              min   avg   max
Connect:        0     1   11
Processing:     0    13   48
Total:          0    13   48

Percentage of the requests served within a certain time (ms)
  50%     12
  95%     23
  99%     29

Verdict

Nginx won on raw throughput by a factor of 2x (15,850 vs 7,524 req/s) with lower and more consistent latency. If you are migrating a reverse proxy setup or serving high-traffic static content, nginx is the right choice. Caddy shines where you want minimal config and zero-effort HTTPS — write a one-line reverse proxy, and Let’s Encrypt is automatic. For a homelab or personal site where throughput is not the bottleneck, Caddy’s ergonomics win.

If you are new to Caddy, my install guide walks through the setup. For nginx-based workflows, see my nginx access log analysis script. This is the latest in my tool comparison series — earlier I compared jq vs Python on a 100K-line log.

The numbers above come from a single run on a 2 vCPU VM. Your mileage will vary by hardware, concurrency, and workload shape. Benchmark your own stack before deciding.