Extend WebRender experiment to Linux

ASSIGNED
Assigned to

Status

task
P2
normal
ASSIGNED
13 days ago
9 days ago

People

(Reporter: tdsmith, Assigned: tdsmith)

Tracking

(Depends on 1 bug, Blocks 2 bugs)

Firefox Tracking Flags

(Not tracked)

Details

(URL)

Brief Description of the request (required):

WebRender eligibility was extended to Intel graphics cards on Linux in bug 1543217. We should assess WR performance on this platform in our prerelease builds in the same way we're assessing WR performance on Windows in bug 1492568.

This should include:

  • shepherding the linked experiment through the experiment launch process
  • producing a dashboard for continuous monitoring

Business purpose for this request (required):

Understand the performance of WebRender on Linux.

Depends on: 1543818

The most recent Linux builds ever had about 60 clients each with gfx_features_wrqualified_status == "available": https://dbc-caf9527b-e073.cloud.databricks.com/#notebook/100696/command/100707

That's pretty slim pickings for an experiment -- we're looking at single-digit percentages of the Windows sample; any per-build metrics are gonna be super noisy.

Priority: -- → P2

Do we know how many of our Nightly users are on Ubuntu 18.04 vs Ubuntu 16.04? Right now the blacklist is setting the minimum Mesa version to 18.2.8.0. On my Ubuntu 16.04 setup I have Mesa 18.0.5.0 which still seems to work ok. If lower the Mesa requirement then we'd presumably pick up all the Ubuntu 16.04 Intel users too which might bring up the population size.

I also remembered that we limit to a bunch of mostly desktop device ids. I'll put up a patch to fix this.

Do we know how many of our Nightly users are on Ubuntu 18.04 vs Ubuntu 16.04?

Not with any precision; kernel version is the only real proxy we have for distribution release.

I added a cell to the bottom of the notebook to look at kernel versions. Looking at https://wiki.ubuntu.com/Kernel/Support I think we expect most 16.04 users to be using 4.4 and most 18.04 users to be using 4.15; 4.15 is about 4x more common than 4.4, so I think lowering the Mesa version threshold probably won't make a huge impact on Nightly.

You need to log in before you can comment on or make changes to this bug.