It's basic benchmark that compares speed of both systems when used in development mode.
Main point is to show that server-side generation of a bundle on request is not necessarily slower than loading modules asynchronously from the browser. It's actually opposite and difference can be significant. See the results section for numbers.
There's no prebuild step, in both cases modules are loaded/bundled on request. In case of AMD, they're loaded asynchronously in a browser. In case of CJS when request occurs they're read from filesystem, bundled and served in one file.
To see fair results (adequate to your development environment) benchmark needs to be installed and launched locally.
You can also check it working at medyk.org:3700, but mind it's backed with basic Node.js server setup (not very efficient for AMD), and due to extra latency, results will be worse than when running same on your localhost.
$ npm install cjs-vs-amd-benchmark
Generate dummy modules (by default it resembles 400 modules tree from existing real world project)
$ npm run setup
You can generate custom number of out of provided modules map:
bin/default-deps-map.json on how it should be constructed.
$ bin/generate --mapPath=path/to/custom/map
Start server (defaults to port 3000)
$ npm start
If you prefer other port, do:
$ bin/start --port=8080
Load the benchmark on corresponding port, e.g. localhost:3000
Following load times were measured on 2008 MBP with local setups of Node.js and Nginx server, each number is average of 5 runs.
Node.js server setup was very basic, no clusters involved, so concurrency was limited (hence not great results for AMD).
Nginx was configured with
Cache-Control: no-cache header, so browser always checks with the server whether new version of module is available.
|CommonJS (via Webmake)||299ms||273ms||N/A||N/A|
|AMD (via RequireJS)||896ms||548ms||367ms||569ms|