curl-benchmark

Simple Python script that runs curl many times, telling it to report time metrics until you press Control+C. These metrics are printed in a colored fashion and then processed to extract average time spent in each phase (DNS lookup, TCP connection, TLS handshake, TTFB...).

Because it's just spawning curl, you can pass arbitrary options to curl by putting them after --. Here we're using --compressed --tcp-nodelay to resemble a browser more closely:

The -r option can be used to hide the raw metrics and only produce the report, and is useful together with -n:

Remember that if you pass options to curl, the URL must go last.

Usage

It's a Python script without dependencies. Just run it.
You need to have Python 3 and cURL installed.
Use -h or --help to get help.

Drawbacks

cURL has lots of options, and some of them significantly affect the time metrics:

• --compressed allows the server to send a compressed version of the asset, reducing content download time in that case.

• Having (and using) HTTP/2 support compiled in cURL can speed up TTFB and download sometimes.

• --false-start is a feature of TLS that allows the handshake to 'complete' in 2RTT instead of the usual 3. So it speeds the handshake by 1RTT.

• --tcp-nodelay supresses queuing of data to be sent, and significantly speeds up the TLS handshake and sometimes the TTFB.

Ideally, if you want to resemble a browser when it comes to time metrics, you should at least use the first three, and a lot of browsers also use the fourth. The problem is that ATM --false-start is not supported unless cURL was compiled with NSS. So you'll have to account for that when interpreting the metrics.

Fixes, improvements, requests

They're welcome! There isn't much code quality yet, though, however you're free to hack it to your needs.