A macro benchmark runner for JavaScript Web apps


Benchpress allows creation and sampling of macro benchmarks to compare performance of real world web applications. The project is built and maintained by the Angular team, and is used to test performance of AngularJS and AngularDart, but is not limited to testing with these frameworks.

$ npm install -g angular-benchpress

Expect frequent breaking changes.

Starting in a project's web app's directory:

  1. Create a directory called "benchmarks" (or some other name if the build step will be provided)
  2. Create a directory in benchmarks/<benchmark-name>
  3. In benchmarks/<benchmark-name>, create a config file called "bp.conf.js" to specify scripts to load in this benchmark
  4. Add a file called "main.html" which is the html that will be interpolated into the benchmark runner template. This is where the markup for the app being tested should live. This is required, although it may be empty.
  5. Create any scripts, html files, or other dependent files in the same folder
  6. Run benchpress build to generate the combined benchmark runner in "benchpress-build/" within the web app
  7. Still in the web app directory, execute benchpress run
  8. Launch Browser (Chrome Canary provides most accurate memory data, See Launching Canary for instructions on testing in Chrome Canary)
  9. Browse to localhost:3339

The benchpress library adds an array to the global bp object called "steps," which is where a benchmark should push benchmark configuration objects. The object should contain a name, which is what the benchmark shows up as in the report, and a fn, which is the function that gets evaluated and timed.

  name: 'Something Expensive',
  description: 'Lengthy description of benchmark...',
  fnfunction() {

Benchpress also exposes an API to manage variables of a test run, useful for comparing test runs under different code conditions. This API is exposed on bp.Variables, and has the following methods and properties:

  • bp.Variables.add({value: 'ngBindOnce'});
  • bp.Variables.addMany([{value: 'ngBindOnce'},{value: 'baseline'}]);
  •'ngBindOnce'); //Select variable by value
  • bp.Variables.selected; //{value:'ngBindOnce'}
  • bp.Variables.variables; //Array of available variables

A variable should be an object with at least a value property, which is a string. Other properties may be added.

Here's how an AngularJS benchmark would incorporate Benchpress variables:

$scope.$watch(function() {return ctrl.benchmarkType}, function(newValoldVal) {;
  value: 'none',
  label: 'none'
$scope.variableStates = bp.Variables.variables;
ctrl.benchmarkType = bp.Variables.selected? bp.Variables.selected.value : undefined;
<div ng-repeat="state in variableStates">{{state.label}}: <input type="radio" name="variableState" ng-model="ctrl.benchmarkType" ng-value="state.value"></div>

See the example in benchmarks/largetable for full reference.

Variables are optional, and are a no-op as far as benchpress is concerned. Benchpress relies on the benchmark code to read and manipulate variable state to change the actual execution of the steps under test. Benchpress provides this API since mosts tests implement variables of some sort, and Benchpress would have a hard time running tests programmatically with variables without some notion of variables.

The default variable to be executed can be provided in the search string of the url using the "variable" parameter name, ie http://localhost:3339/benchpress-build/largetable?variable=ngBindOnce.

There is one variable state set for all steps at any given time.

There are no sophisticated mechanisms for preparing or cleaning up after tests (yet). A benchmark should add a step before or after the real test in order to do test setup or cleanup. All steps will show up in reports.

Each benchmark directory should contain a file named bp.conf.js, which tells benchpress how to prepare the benchmark at build-time.

Example benchpress config:

module.exports = function(config) {
    //Ordered list of scripts to be appended to head of document 
    scripts: [{
      id: 'angular', //optional, allows overriding script at runtime by providing ?angular=/some/path, 
      src: '../../../build/angular.js' //relative path to library from runtime benchmark location 

The CLI has three commands:

$ benchpress build --build-path=optional/path
$ benchpress run --build-path=optional/path //Starts serving cwd at :3339. Will redirect '/' to build-path
$ benchpress launch_chrome //Launches Chrome Canary as described below

For Mac and Linux computers, a utility script is included to launch Chrome Canary with special flags to allow manual garbage collection, as well as high resolution memory reporting. Unless Chrome Canary is used, these features are not available, and reports will be lacking information. Samples will also have more outliers with more expensive test runs because garbage collection timing is left up to the VM.

This launches Chrome Canary in Incognito Mode for more accurate testing.

$ benchpress launch_chrome

After opening the benchmark in the browser as described in Creating Benchmarks, the test execution may be configured in two ways:

  1. Number of samples to collect
  2. How many test cycles to run

The number of samples tells benchpress "analyze the most recent n samples for reporting." If the number of samples is 20, and a user runs a loop 99 times, the last 20 samples are the only ones that are calculated in the reports. This value is controlled by a text input at the top of the screen, which is set to 20 by default.

The number of times a test cycle executes is set by pressing the button representing how many cycles should be performed. Options are:

  • Loop: Keep running until the loop is paused
  • Once: Run one cycle (note that the samples will still be honored, pressing once 100 times will still collect the number of samples specified in the text input)
  • 25x: Run 25 cycles and stop, still honoring the specified number of samples to collect