Add-on Power Tests

This project is a stupid simple test harness to test the affects of add-ons on Firefox's power consumption using Intel's Power Gadget. It consists of two components: a script to setup and launch Firefox, and an add-on to control it.

runner.sh

This script provisions Firefox with any number of add-ons, launches it via Intel's Power Gadget power recorder, and then stores the results. When run with arguments, the skeleton profile profile/ is copied to a new directory, and all arguments are assumed to be extension XPIs, and are installed to it. On subsequent runs with no arguments, the previously instantiated profile is reused.

The following variables affect the script's operation:

$FIREFOX

The path to the Firefox executable. Default: C:/Program Files (x86)/Mozilla Firefox/firefox.exe

$POWERLOG

The path to the Intel Power Gadget PowerLog executable. Default: C:/Program Files/Intel/Power Gadget 2.6/PowerLog.exe

$SKELETON_PROFILE

The path to the skeleton profile from which to initialize the test profile. Default: ./profile

$POWERLOG_OUTPUT

The path to the power log output file. If omitted, the log will not be saved. If present, $POWERLOG_JSON will be set to an appropriate value for the harness add-on.

The script is adapted to run on Windows in a MinGW or Cygwin environment, but can be adapted to run on any platform supported by Intel's Power Gadget.

The harness add-on

The addon/ directory contains a simple test harness add-on which is intended to be installed via the runner.sh script. On the first run after installation, it waits for the first browser window to open and then quits after a short delay. This allows add-ons to complete first-run initialization without impacting power tests on subsequent runs.

On second and subsequent runs, the add-on waits for the first browser window to open, opens a series of URLs in tabs, and then cleans up and quits after a 30 second delay. If the $POWERLOG_JSON environment variable is set, it also dumps timing information for the various actions it completes, and several memory usage samples to the file named therein.

The testing process

An appropriate skeleton profile need to be created and saved to the patch indicated by $SKELETON_PROFILE. At the very least, the profile needs to include the following setting changes:

• Update mechanisms need to be disabled, including app, extension, add-on metadata, safe browsing database, blocklist, and sync updates.
• Firefox Health Report and Telemetry reporting should be disabled.
• The startup.homepage_override_url preference should be cleared.

When running on Windows, all unnecessary services should be disabled, and at least the following changes should be made:

• Disable the disk defragmenter, search indexer, screen saver, Windows updates, Windows Firewall, Windows Defender, any anti-virus services.
• Disable the Firefox update service.
• Change power settings to disable display dimming, adaptive brightness, suspend/hibernate. Select the 'Balanced' power plan.

The test process for a given add-on should be as follows:

1. Invoke runner.sh with the path to the XPI of the add-on to be tested and the path to the harness add-on.
2. Set the $POWERLOG_OUTPUT variable to the file where the PowerLog CSV should be saved. Invoke runner.sh again without arguments.
3. Repeat the above several times, with non-clashing values for $POWERLOG_OUTPUT each run, and use the run with the median cumulative power usage for the basis of analysis.