Skip to content
This repository

HTTPS clone URL

Subversion checkout URL

You can clone with HTTPS or Subversion.

Download ZIP

Compare load time of CommonJS and AMD style modules

branch: master

Fetching latest commit…

Octocat-spinner-32-eaf2f5

Cannot retrieve the latest commit at this time

Octocat-spinner-32 bin
Octocat-spinner-32 cjs-modules
Octocat-spinner-32 public
Octocat-spinner-32 scripts
Octocat-spinner-32 .gitignore
Octocat-spinner-32 .lint
Octocat-spinner-32 CHANGES
Octocat-spinner-32 LICENCE
Octocat-spinner-32 README.md
Octocat-spinner-32 package.json
Octocat-spinner-32 server.js
Octocat-spinner-32 webmake-middleware.js
README.md

CJS/AMD load time benchmark

Compare load times of two module systems.

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.

AMD modules are loaded with RequireJS and CommonJS modules are bundled on request with Webmake.

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.

Installation

Install package:

$ 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:
See 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

Results

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.

Node.js v0.10.13
Google Chrome
Node.js v0.10.13
Firefox
Nginx v1.0.10
Google Chrome
Nginx v1.0.10
Firefox
CommonJS (via Webmake) 299ms 273ms N/A N/A
AMD (via RequireJS) 896ms 548ms 367ms 569ms
Something went wrong with that request. Please try again.