Page MenuHomePhabricator

Investigate: Chrome 111 seems to increase the performance on LCP and FCP on first view desktop
Closed, ResolvedPublic

Assigned To
Authored By
Peter
Mar 13 2023, 7:04 AM
Referenced Files
F36910690: direct.jpg
Mar 14 2023, 7:39 AM
F36910688: barack111.jpg
Mar 14 2023, 7:39 AM
F36910685: Screenshot 2023-03-14 at 08.31.57.png
Mar 14 2023, 7:39 AM
F36910684: Screenshot 2023-03-14 at 08.32.10.png
Mar 14 2023, 7:39 AM
F36908677: Screenshot 2023-03-13 at 08.00.00.png
Mar 13 2023, 7:04 AM
F36908675: Screenshot 2023-03-13 at 08.00.10.png
Mar 13 2023, 7:04 AM
F36908671: Screenshot 2023-03-13 at 07.58.41.png
Mar 13 2023, 7:04 AM
Subscribers

Description

On the new bare metal server I've pushed Chrome 111 last week and I could see an increase in performance for all pages that we test on first view for desktop, This is what it looks like on Barack Obama page:

Screenshot 2023-03-13 at 07.58.41.png (1×2 px, 451 KB)

Looking at the rendering before:

Screenshot 2023-03-13 at 08.00.10.png (842×2 px, 424 KB)

And after:

Screenshot 2023-03-13 at 08.00.00.png (922×2 px, 451 KB)

I've checked on our Android phones to see if this has any effect on Android too but 111 hasn't been rolled out there yet.

There's a couple of things that I want to do:

  1. Check with our RUM data If I can spot the same there (I checked this weekend but too few Chrome had been updated then)
  2. Update our tests on AWS. We are behind there running on old Chrome 105. Let update to first 110 and then after a couple of runs 111.
  3. Check on our mobile phones if we can see the same thing there.
  4. Ask the Chrome team about if they know what's been changes. I briefly looked at the raw change log between 110->111 but could see anything but I need to dig deeper.

Event Timeline

Change 897787 had a related patch set uploaded (by Phedenskog; author: Phedenskog):

[performance/synthetic-monitoring-tests@master] Updating to sitespeed.io new beta to Chrome 110.

https://gerrit.wikimedia.org/r/897787

Change 897787 merged by jenkins-bot:

[performance/synthetic-monitoring-tests@master] Updating to sitespeed.io new beta to Chrome 110.

https://gerrit.wikimedia.org/r/897787

I've pushed the changes so we run Chrome 110 in all our testing and by tonight I'll push 111 and then we can see tomorrow what it looks like.

Change 897855 had a related patch set uploaded (by Phedenskog; author: Phedenskog):

[performance/synthetic-monitoring-tests@master] Use container that uses Chrome 111.

https://gerrit.wikimedia.org/r/897855

Change 897855 merged by jenkins-bot:

[performance/synthetic-monitoring-tests@master] Use container that uses Chrome 111.

https://gerrit.wikimedia.org/r/897855

First let's have a look at our RUM data for First Contentful Paint for 110 and 111 looking at p75:

Screenshot 2023-03-14 at 08.32.10.png (1×2 px, 440 KB)

Screenshot 2023-03-14 at 08.31.57.png (1×2 px, 413 KB)

No there's no difference there,

Let's look at the AWS tests. Here are running with WebPageReplay, testing Barack Obama:

barack111.jpg (1×2 px, 331 KB)

And the direct tests:

direct.jpg (1×2 px, 268 KB)

No changes there either. Interesting. On the bare metal server that runs on a slow CPU we could see a big change, also on another bare metal server that runs on faster speed I could see the same. Let me look at the Chrome trace log and compare before and after and see if I can see anything there.

Change 898733 had a related patch set uploaded (by Phedenskog; author: Phedenskog):

[performance/synthetic-monitoring-tests@master] Stop disabling PaintHoldingCrossOrigin.

https://gerrit.wikimedia.org/r/898733

Change 898733 merged by jenkins-bot:

[performance/synthetic-monitoring-tests@master] Stop disabling PaintHoldingCrossOrigin.

https://gerrit.wikimedia.org/r/898733

So it turns out I was running different configuration on AWS and the bare metal. On AWS we have disabled PaintHoldingCrossOrigin, but that is fixed now, lets see if we can see any result.

Interesting: T299886 and https://bugs.chromium.org/p/chromium/issues/detail?id=1291502#c63

This is rolling out now -- part of the fix is included by default in M111, and the remainder is ramping up to 100% over the next few days.

I'll mark this as fixed once the fixed behaviour is set to be the default in ToT Chrome as well, but you should be seeing results in 111 already.

That can also affect the metric.