Report of yesterday's "web performance meetup" in Paris

[Julien Wajsberg]

Hi,

yesterday I went to the first edition of the web performance meetup in 3
years. This is a regular meetup with most web performance experts in
Paris. Usually very interesting talks, and then very interesting
discussions.

The talks this time were about:

* what changed in 3 years from Jean-Pierre Vincent.
o more mobile, more https
o most mobile phones are shitty
o bandwidth more or less similar -- at least in France
o responsive images
o twice as much data, especially javascript => challenge is how to
execute faster big chunks of JS, in a DOM tree twice bigger
o cool tools, eg www.sitespeed.io, http://yellowlab.tools/, and of
course still webpagetest
o explosion of 3rd-party scripts usage -- cf last talk
* feedback about HTTP/2 from Fasterize CEO Stéphane Rios
o not a magic bullet
o needs to check existing practices, especially domain sharding,
but also concatenation
o push is difficult to manage, needs to take cache into account
* hidden costs of 3rd-party scripts from JP Vincent
o analytics, social buttons, widgets, and ads
o make websites slower, but the website's image is hurt, not the
3rd-party's
o risk of industrial piracy
o SPOF: 3rd-party servers might not respond sometimes: DDos,
network errors, China's Firewall, the server doesn't exist
anymore, Firefox Tracking Protection
o they need to be inserted asynchronously
o CPU usage and track usage per hostname
o 5 share buttons -> +400 KB, +22 requests, +CPU, +2s loading,
+400ms first paint
o widgets help track your users
o your 3rd-party scripts loads other 3rd-party scripts, which
loads... (see lightbeam) -> data piracy is possible from these
transitive scripts we don't know
o adblock users engage 50% more on websites


Then I had a discussion with several people, asking them how we could
help them. Turns out that for these experts the biggest demand is
automatic testing and monitoring of the performance.
For example it's especially difficult to know the time spent on the
various rendering steps: CSS application, layout, display list,
painting. It's also difficult to measure 3rd-party scripts' impact.
They also use Chrome Dev tools' timeline which can export a JSON, and
then they use custom scripts to get this information.


Next meetup will likely happen at Mozilla Paris office and I already
submitted a talk about our performance tools :)

--
Julien

[Schalk Neethling]

Thanks for sending this out, some really insightful pieces in there.

On Thu, Mar 23, 2017 at 4:02 PM, Julien Wajsberg <jwaj...@mozilla.com>
wrote:

> _______________________________________________
> dev-developer-tools mailing list
> dev-devel...@lists.mozilla.org
> https://lists.mozilla.org/listinfo/dev-developer-tools
>
>


--
Kind Regards,
Schalk Neethling
Senior Front-End Engineer
Mozilla ::-::

[Harald Kirschner]

> They also use Chrome Dev tools' timeline which can export a JSON, and
then they use custom scripts to get this information.

Did people talk about automation there? Ember & FB [1] have similar tools using WebDriver which Firefox doesn't support (bug 1250290). Webpagetest and sitespeed.io also show profile breakdown, but again only for Chrome but that could be done for Firefox if somebody invests some time to contribute [2].

[1]: https://code.facebook.com/posts/1368798036482517/browserlab-automated-regression-detection-for-the-web/
[2]: https://github.com/WPO-Foundation/webpagetest/issues/744

[Julien Wajsberg]

Le 24/03/2017 à 18:38, Harald Kirschner a écrit :
>> They also use Chrome Dev tools' timeline which can export a JSON, and
> then they use custom scripts to get this information.
>

> Did people talk about automation there?

Yep that's what I called "automatic testing" :)

> Ember & FB [1] have similar tools using WebDriver which Firefox doesn't support (bug 1250290). Webpagetest and sitespeed.io also show profile breakdown, but again only for Chrome but that could be done for Firefox if somebody invests some time to contribute [2].

Yep, they especially use these 2 tools.
I agree with you. Lacking this support, developers target being
performant in Chrome only...

>
> [1]: https://code.facebook.com/posts/1368798036482517/browserlab-automated-regression-detection-for-the-web/
> [2]: https://github.com/WPO-Foundation/webpagetest/issues/744
>
> On Thursday, March 23, 2017 at 7:02:33 AM UTC-7, Julien Wajsberg wrote: