Webview performance degraded after upgrading it from version 58 to 85 in android 8.1. Why ?

[Xuening Deng]

I have a smart device equipped with android 8.1 system. I am planning to optimize webview performance for a better user experience, So I decided to upgrade webview. I compiled a webview 85 from source code and installed it on my smart device, but benchmark tests  result shows that performance degraded. Why the higher version of webview(85) inferior to that of a lower version(58) ?

Test tools:

Speedometer2.0
Octane
Sunspider

[Torne (Richard Coles)]

On Fri, 30 Oct 2020 at 13:39, Xuening Deng <dengxue...@gmail.com> wrote:

I have a smart device equipped with android 8.1 system. I am planning to optimize webview performance for a better user experience, So I decided to upgrade webview. I compiled a webview 85 from source code and installed it on my smart device, but benchmark tests  result shows that performance degraded. Why the higher version of webview(85) inferior to that of a lower version(58) ?

What are the actual results? How much worse is it?

Were you using the prebuilt webview 58 before, or a version you had previously compiled yourself? What build settings did you use to build it?

Have you tried any other versions in between? (that's a very, very large version range)

 

Test tools:

Speedometer2.0
Octane
Sunspider

--
--
Chromium Discussion mailing list: chromium...@chromium.org
View archives, change email options, or unsubscribe:
http://groups.google.com/a/chromium.org/group/chromium-discuss

[Xuening Deng]

Sorry, the information is not clear.The details of the testing results are as follow:

Test tools:

Octane
Sunspider

Speedometer2.0

Test results:

version 58(prebuilt)

• octane score: 2636
• sunspider time-consuming: 1454ms
• Speedometer2.0: 6.04 times/min
  

version 85:

• octane score: 2272 
• sunspider time-consuming: 1984ms
• Speedometer2.0: 7.15 times/min

Description of testing results:

• Octane: The higher the score, the better the efficiency of JS execution.
• sunspider:  The shorter the time, the better the efficiency of JS execution.
• Speedometer2.0: The more executions, the better the performance. Depends on the speed of the DOM APIs, the JavaScript engine, CSS style resolution, layout, and other technologies.

Build settings of version 85:

gn gen --args='target_os="android" target_cpu="arm" proprietary_codecs=true ffmpeg_branding="Chrome" is_debug=false is_official_build=true is_chrome_branded=false use_official_google_api_keys=false exclude_unwind_tables=true enable_resource_whitelist_generation=true enable_remoting=true symbol_level=1 ' out/Default

autoninja -C out/Default system_webview_apk

I havent't tried the intermediate version, because I don't know which version is the best. And I think the higher the version, the better the performance, but test results show it is not.

Could you please give me some advice ? Thanks a lot !

[Torne (Richard Coles)]

Those build options seem reasonable - I would suggest you try to roughly bisect it by trying different major versions in the range between.

Is the device you are testing on actually 32-bit?

[Xuening Deng]

Thanks for your advice. The device is actually 32-bit. I'm still confused about performance degradation. Is there anybody know the deeper reason ?

[Torne (Richard Coles)]

On Wed, 4 Nov 2020 at 06:59, Xuening Deng <dengxue...@gmail.com> wrote:

Thanks for your advice. The device is actually 32-bit. I'm still confused about performance degradation. Is there anybody know the deeper reason ?

We're not aware of a regression, but the range you have is so large that actually investigating it would be difficult. You will need to try other versions and see if you can narrow down the range where the regression occurred.

[Xuening Deng]

Thanks. This is a bit too time consuming, I will give a try.