- From: <steve@steveclaflin.com>
- Date: Fri, 06 Mar 2015 09:59:18 -0600
- To: Simon Pieters <simonp@opera.com>
- Cc: Paul Deschamps <pdescham49@gmail.com>, Jason Grigsby <jason@cloudfour.com>, public-respimg@w3.org
It amazes me how things circle around - this reminds me of the old Netscape lowsrc attribute, which I believe was an attempt to resolve the same basic issue. Too bad that kicks off an additional request. The whole preloading thing is a bit of a double-edged sword. There are a number of pages that really hang some browsers for me (nhl.com in Chrome comes to mind), for which I believe the issue is way too many remote dynamically generated images (aka advertisements) all trying to load at once. On 2015-03-06 04:28, Simon Pieters wrote: > On Thu, 05 Mar 2015 16:30:23 +0100, Jason Grigsby <jason@cloudfour.com> > wrote: > >> In Bruce's article, he quotes Steve Souders as saying that the >> preloader is >> the "the single biggest performance improvement browsers have ever >> made". >> >> In the link I provided, Andy Davies talks about how Google saw a 20% >> and >> Firefox a 19% increase in average page speed after implementing the >> preloader. > > To be fair, I think those numbers are not entirely relevant for > speculatively loading images per se. The biggest performance > improvement is from speculatively loading other scripts and > stylesheets while the HTML parser is blocked on loading a script. I'm > not aware of comparisons of not speculatively loading only images. > > But even if you don't load images speculatively, the browser could > still start loading images when the real parser sees the <img> which > can happen before CSS has loaded, as in my example in > https://lists.w3.org/Archives/Public/public-respimg/2015Mar/0043.html > (there are no scripts in the example and thus nothing is loaded > speculatively). > > If you were to defer image loading until layout is known, it needs to > be deferred in both the speculative parser *and* when the real parser > creates the element. Browsers have never waited with loading images > until layout is known. As can be seen in my example, the best-case > scenario would be that images pop in 1 RTT after first paint if > loading images were to wait for layout. > > Put another way, the perf regression of the proposal would probably > not be 20% page loading time, but instead having first paint be > without any images and have images come in 1RTT later.
Received on Friday, 6 March 2015 15:48:52 UTC