Issue metadata
Sign in to add a comment
|
Initial layout perf on Facebook
Reported by
bmau...@fb.com,
Apr 2 2016
|
||||||||||||||||||||||
Issue description<sorry most of this might not make sense, it's mostly a copy/paste from an email thread> I’ve been looking at a number of our traces from production and I’ve noticed that a large gap of time (100s of ms in some cases) between the last <script> that is a part of the head and the first one that is a part of the body. During this gap we often see image requests go out for images that are a background-image set by a stylesheet. Based on this I was wondering if that gap might be caused by the initial layout and styling of the document. I used webpage test to gather a trace of https://www.facebook.com/barackobama/ as a logged out user. In this trace I can see a pattern that matches what I see in my production traces — it takes 175 ms to do an initial style & layout of the document. The dom really isn’t that big at this point — UpdateLayoutTree says there are 917 elements at this point in the page. WPT results: http://www.webpagetest.org/result/160401_YC_1B32/ http://www.webpagetest.org/export.php?bodies=1&pretty=1&test=160401_YC_1B32 I attached a zipped HTML directory with additional JS/CSS. I used Chrome’s “export to HTML” and I had to manually remove crossorigin=anonymous from the links (I’m assuming some security feature is banning that type of load from local disk). I was able to reproduce the layout jank with this. It’s not exactly minimal, but it’s at least a start. This seems to reproduce on https://www.facebook.com/legal/terms, which should be a much simpler page. Here’s a WPT render of that page http://www.webpagetest.org/result/160402_TJ_VPZ Attached are three test cases: 1) Obama's logged out fb page 2) A reduced version of obama's FB page 3) Our ToS, reduced
,
Jul 25 2016
|
|||||||||||||||||||||||
►
Sign in to add a comment |
|||||||||||||||||||||||
Comment 1 by cbiesin...@chromium.org
, Jul 23 2016