Open Bug 1556534 Opened 8 months ago Updated 16 days ago

WebRender vs. Non-WebRender Experiment for Fenix release

Categories

(Data Science :: Experiment Collaboration, task)

Other
Android
task
Not set
Points:
3

Tracking

(Not tracked)

People

(Reporter: jbonisteel, Unassigned)

References

(Blocks 2 open bugs)

Details

Brief Description of the request (required):

We have been working on getting WebRender on Android targeting specific devices (Pixel 2, to start) and we would like to be able compare the impact of WR vs. non WR, similar to what we did for desktop. We are figuring out the telemetry here: https://bugzilla.mozilla.org/show_bug.cgi?id=1546149

Business purpose for this request (required):

To assess performance impact and ensure no regressions happen for WebRender on Android

Requested timelines for the request or how this fits into roadmaps or critical decisions (required):

We'd like to try turning WR on in Nightly 70 for Pixel 2 (so around July time-frame) so we can do additional testing.

Links to any assets (e.g Start of a PHD, BRD; any document that helps describe the project):

Name of Data Scientist (If Applicable):

Tim Smith has helped with previous WebRender work

Please note if it is found that not enough information has been given this will delay the triage of this request.

Summary: WebRender vs. Non-WebRender Experiment for Geckoview → WebRender vs. Non-WebRender Experiment for Geckoview (nightly)

I'll hold this bug in my backlog but it looks like it's blocked on geckoview telemetry integration.

Assignee: nobody → tdsmith
Blocks: 1522903
Status: NEW → ASSIGNED
data-science-status: --- → Blocked
Points: --- → 3
Depends on: 1546149
Priority: -- → P3

Clearing my assignment for retriage on Monday and cc'ing Kris; this looks related to the email you sent me today.

Assignee: tdsmith → nobody
Status: ASSIGNED → NEW
data-science-status: Blocked → ---
Priority: P3 → --

This would be our second, I think, experiment in Firefox Preview; bug 1574618 follows the first.

We're currently on track to enable Pixel2 and Pixel3 models for Firefox 72, which would mean we would want the experiment to run when Fenix 4.0 is released (~Jan 20).

Pixel 2/3 is enabled in FF 72 beta and is on track to be shipped with the Fenix 3.1 release (Planned release Jan 20).
https://bugzilla.mozilla.org/show_bug.cgi?id=1600133

There is one more blocker that is causing telemetry values to appear as null. Fix is landing on nightly and once it is verfied I will request uplift to beta.
https://bugzilla.mozilla.org/show_bug.cgi?id=1601091

Depends on: 1601091, 1600133
Depends on: 1604862

Hey Tim,

I think everything that blocked this experiment from Gfx side is now completed and uplifted to beta.
Is there anything else you need from us or are the next steps for the data science team?

Flags: needinfo?(tdsmith)
Summary: WebRender vs. Non-WebRender Experiment for Geckoview (nightly) → WebRender vs. Non-WebRender Experiment for Fenix release

Just to clarify the request (as I can't change the initial comment).

This experiment would be for the Fenix 3.1 release which is the first shipping with GV 72. We would want to use it to control our rollout of webrender on Pixel 2/3 devices and ensure stability, quality and performance are satisfactory.

To recap a conversation with Kris in slack/#gfx:

  • In GV 72, WebRender will be enabled by default on the Pixels
  • gfx.webrender.force-disabled is a pref that can be used to disable WebRender for an A/B test
  • https://github.com/mozilla-mobile/fenix/blob/master/docs/experiments.md describes how to implement an experiment in Fenix, since we can't flip prefs directly with the android-experiments library; support is available in slack/#mobile-experiments.
  • Kris will look into the implementation work and open an Experimenter ticket.
Flags: needinfo?(tdsmith)
See Also: → 1605454
You need to log in before you can comment on or make changes to this bug.