Telemetry: Run Benchmarks Locally

Set Up

If you just want Telemetry, cloning catapult repository should be enough. If you also want the Chrome benchmarks built with Telemetry, get the latest Chromium checkout. If you‘re running on Mac OS X, you’re all set! For Windows, Linux, Android, or ChromeOS, 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, you need to install the libstdc++6:i386 package.

Android

Running on Android is supported with a Linux host. Windows and Mac OS X are 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. Telemetry only supports devices with ADB Daemon running 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. If you are unable to install “userdebug” build of Android, you can try running benchmarks with --compatibility-mode=dont-require-rooted-device switch, however this configuration may not be supported and you may run into errors.
  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>.

ChromeOS

See Running Telemetry on ChromeOS.

Benchmark Commands

Telemetry benchmarks can be run with 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:

src/tools/perf/run_benchmark blink_perf.css --browser=stable

Running on another browser

To list available browsers, use:

src/tools/perf/run_benchmark --browser=list

For ease of use, you can use default system browsers on desktop:

src/tools/perf/run_benchmark blink_perf.css --browser=system

and on Android:

src/tools/perf/run_benchmark blink_perf.css --browser=android-system-chrome

If you‘re running telemetry from within a Chromium checkout, the release and debug browsers are what’s built in out/Release and out/Debug, respectively.

To run a specific browser executable:

src/tools/perf/run_benchmark blink_perf.css --browser=exact --browser-executable=/path/to/binary

To run on a Chromebook:

src/tools/perf/run_benchmark blink_perf.css --browser=cros-chrome --remote=[ip_address]

Options

To see all options, run:

src/tools/perf/run_benchmark run --help

Use --pageset-repeat to run the benchmark repeatedly. For example:

src/tools/perf/run_benchmark blink_perf.css --pageset-repeat=30

Use --run-abridged-story-set to run a shortened version of a benchmark with a representative subset of the stories included (Note that some benchmarks do not have an abridged version yet. Instructions for abridging a benchmark are here). Example:

src/tools/perf/run_benchmark rendering.desktop --run-abridged-story-set

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

src/tools/perf/run_benchmark blink_perf.css --reset-results --results-label="foo"

Comparing Two Runs

src/tools/perf/run_benchmark some_benchmark --browser-executable=path/to/version/1 --reset-results --results-label="Version 1"

src/tools/perf/run_benchmark some_benchmark --browser-executable=path/to/version/2 --results-label="Version 2"

The results will be written to in the results.html file in the same location of the run_benchmark script.