1367205 - (QF-Polymer) [meta] Very poor Polymer performance compared to Chrome (Web Components v0)

Status: RESOLVED INCOMPLETE

(Core :: DOM: Core & HTML, defect, P2)

Description

[Sean Stangl [:sstangl]]

Reported by a user on https://www.reddit.com/r/firefox/comments/6cwn5s/gamingyoutubecom_performance_in_firefox_vs_chrome/, it appears that http://gaming.youtube.com takes about twice as long to load in Firefox Nightly as it does to load in Chrome Stable.

Taking a profile, https://perfht.ml/2q8dMXv, a significant chunk of time (~1s of 2s) is spent in gaming_polymer.js, which is their deployment of the Polymer framework.

An even more dramatic difference can be seen if you visit the Polymer homepage at https://www.polymer-project.org/. Chrome loads that effectively instantly, but Firefox takes between 7-10 seconds.

Polymer is a Google library, and it seems that they did not build it with Firefox performance in mind, but we nevertheless must improve the situation.

Comment 1

[Ben Kelly [:bkelly, not reviewing]]

Polymer has to polyfil web components on FF.  Chrome implements the web components APIs natively so probably a lot of savings come from that.

Comment 2

[Sean Stangl [:sstangl]]

There is a benchmark made by the Polymer team for an older version (0.8; current 1.9) available at http://polymerlabs.github.io/benchmarks/synth/medium-list/runner.html.

Running that in Firefox Nightly versus Chrome Stable gives the following timings:

> polymer/app.html:
> Chrome: 8.9ms; Firefox: 1551.9ms

> polymer-0.8-03242015/app-shadow.html
> Chrome: 225.7ms; Firefox: 1093.7ms

> polymer-0.8-03242015/app.html
> Chrome: 253.0ms; Firefox: 307.3ms

> polymer-0.8-03242015-class/app.html
> Chrome: 208.1ms; Firefox: 304.9ms

Comment 3

[Caspy7]

The original reporter of the Youtube Gaming performance issue left a comment that may be worth sharing.

> Hopefully it will be fixed before polymer is on youtube.com as well, which is currently available as an option on youtube.com/new and performance is pretty much the same as with gaming.youtube.com.

So, if Polymer is coming to Youtube, we *really* want to get on top of it. Presumably it will be spreading to other websites as well.

Comment 4

[Naveed Ihsanullah [:naveed]]

Do we have a timeframe for webcomponents? Maybe a subset implementation gets us far enough?

Comment 5

[Andrew Overholt [:overholt]]

(In reply to Naveed Ihsanullah [:naveed] from comment #4)
> Do we have a timeframe for webcomponents? Maybe a subset implementation gets
> us far enough?

Looking like end of 2017 for Custom Elements and currently unknown for Shadow DOM.

Comment 6

[Robert Johnston]

Possibly (probably) related. Latest FireFox nightly (57.0a1 (2017-08-19) (64-bit)) does not load YouTube's "New" implementation at all, reporting "TypeError: Polymer.DomModule.import is not a function". This is with webcomponents and custom elements enabled in about:config.

The main polymer project site DOES work, but reports 3 instances of the same error.

Comment 7

[Olli Pettay [:smaug][[e-mail]]]

The pref to enable webcomponents isn't supported configuration.

Comment 8

[Robert Johnston]

Okay, but the behaviour is the same whether webcomponents are enabled or not.

Comment 9

[Caspy7]

It's notable for this bug that Google is now turning on the new Polymer-based Youtube interface for many (all?) of their users.

A recent [healthily upvoted] post on reddit has several comments complaining about very poor performance with the new UI in Firefox 55.
https://www.reddit.com/r/firefox/comments/70fy9b/new_youtube_interface_is_hogging_all_resources_in/

Comment 10

[Robert Johnston]

Incidentally, my issue was being caused by a popup blocker, and is not relevant.

Comment 11

[Andrew Overholt [:overholt]]

(In reply to Caspy7 from comment #9)
> It's notable for this bug that Google is now turning on the new
> Polymer-based Youtube interface for many (all?) of their users.
> 
> A recent [healthily upvoted] post on reddit has several comments complaining
> about very poor performance with the new UI in Firefox 55.
> https://www.reddit.com/r/firefox/comments/70fy9b/
> new_youtube_interface_is_hogging_all_resources_in/

I can't reproduce (Task Manager on Windows 10 on a Surface Book peaked around 30% for Nightly). Can someone who can capture a profile (https://developer.mozilla.org/en-US/docs/Mozilla/Performance/Reporting_a_Performance_Problem)?

Comment 12

[Jan Hutaø]

OK, not sure if what I see (firefox-59.0.1-1.fc27.x86_64 eating a lot of CPU when on youtube.com) is what is discussed in this BZ, but this is mine perf profile: https://perfht.ml/2J6g0MF

Comment 13

[Andrew Overholt [:overholt]]

(In reply to Jan Hutaø from comment #12)
> OK, not sure if what I see (firefox-59.0.1-1.fc27.x86_64 eating a lot of CPU
> when on youtube.com) is what is discussed in this BZ, but this is mine perf
> profile: https://perfht.ml/2J6g0MF

Thanks! Looks like you would benefit from OMTP (bug 1432531 is for enabling it on Linux).

Comment 14

[Andrew Overholt [:overholt]]

We're going to ship Custom Elements and Shadow DOM v1 and not v0 which is what many existing Polymer-based sites use. Hence me breaking the dependency here.

Comment 15

[Chris Peterson [:cpeterson]]

Linking to other Polymer performance bugs

Comment 16

[ExE Boss]

I came across a Custom Elements v0 polyfill which uses the browser’s native Custom Elements v1 implementation under the hood: https://github.com/WebReflection/ce-v0

This is better than the polyfill used by Polymer v1 (the version used on YouTube¹) which implements Custom Elements v0 from scratch regardless if the browser supports Custom Elements v1.

¹ Determined by checking the value of `window.Polymer.version` in the browser console.

Comment 17

[ExE Boss]

My idea is that the polyfill could potentially be used as a reference for implementing a v0 wrapper around the v1 API implemented in Firefox.

Comment 18

[Dave Hunt [:davehunt] [he/him] ⌚BST]

Quantum Flow was a cross-team browser responsiveness improvement effort that ran throughout 2017 and culminated in the release of Firefox Quantum in November of that year. The dependencies of this meta bug are likely still valid and offer potential to improve the performance of Firefox even further, however we are no longer tracking these under the "Quantum Flow" effort. For that reason, this bug is being resolved as incomplete. You can read more about the realese of Firefox Quantum in this article and learn about our latest performance triage process here.