Telemetry: Run Benchmarks Locally

Contents

  1. 1 Set Up
    1. 1.1 Windows
    2. 1.2 Android
    3. 1.3 Chrome OS
  2. 2 Benchmark Commands
  3. 3 Telemetry benchmarks can be run with run_benchmark.
    1. 3.1 Running a benchmark
    2. 3.2 Running on another browser
    3. 3.3 Options
    4. 3.4 Comparing Two Runs

Set Up

If you don't have a Chromium checkout, download the latest Telemetry archive. Unzip the archive. If you're running on Mac OS X, you're all set! For Windows, Linux, Android, or Chrome OS, read on.

Windows

Some benchmarks require you to have pywin32. Be sure to install a version that matches the version and bitness of the Python you have installed.

Linux

Telemetry on Linux tries to scan for attached Android devices with ADB. The included adb binary is 32-bit. On 64-bit machines, one needs to install the libstdc++6:i386 package.

Android

Running on Android is supported with a Linux or Mac OS X host. Windows is not yet supported. There are also a few additional steps to set up:

  1. Telemetry requires ADB. If you're running from the zip archive, ADB is already included. But if you're running with a Chromium checkout, ensure your .gclient file contains target_os = ['android'], then resync your code.

  2. If running from an OS X host, you need to run ADB as root. First, you need to install a "userdebug" build of Android on your device. Then run adb root. Sometimes you may also need to run adb remount.

  3. Enable debugging over USB on your device.

  4. You can get the name of your device with adb devices and use it with Telemetry via --device=<device_name>

Chrome OS

See Running Telemetry on Chrome OS.

Benchmark Commands

Telemetry benchmarks can be run with run_benchmark.

In the Telemetry zip archive, this located at telemetry/run_benchmark.
In the Chromium source tree, this is located at src/tools/perf/run_benchmark.

Running a benchmark

List the available benchmarks with:

$ telemetry/run_benchmark list

Here's an example for running a particular benchmark:

$ telemetry/run_benchmark --browser=canary smoothness.top_25

Running on another browser

To list available browsers:

$ telemetry/run_benchmark --browser=list

If you're running telemetry from chromium src, the release and debug browsers are what's built in out/Release and out/Debug, respectively. If you are trying to run on Android, you should see an entry similar to android-jb-system-chrome in this list. If it's not there, the device is not set up correctly.

To run a specific browser executable:

$ telemetry/run_benchmark --browser=exact --browser-executable=path_to_binary

To run on a Chromebook:

$ telemetry/run_benchmark --browser=cros-chrome --remote=ip_address

Options

To see all options, run:

$ telemetry/run_benchmark run --help

Use --pageset-repeat to run the test repeatedly.

$ telemetry/run_benchmark smoothness.top_25 --pageset-repeat=30

If you want to generate HTML Results, you can do this locally by using the parameter --reset-results and --results-label="foo"

$ telemetry/run_benchmark smoothness.top_25 --reset-results --results-label="foo"

Comparing Two Runs

$ telemetry/run_benchmark some_test --browser-executable=path/to/version/1 --reset-results --results-label="Version 1"
$ telemetry/run_benchmark some_test --browser-executable=path/to/version/2 --results-label="Version 2"

The results will be written to in src/tools/perf/results.html
Comments