Yes, I know this was a while ago, but I finally made the time to listen through this entire presentation:
Kyle Scholz and Yaron Friedman are front-end developers for Google Search, focused on one thing: Make Google Search faster. The main bullet points to me are:
- Remove unnecessary tags & whitespace:
- Google Search HTML weighs just over 10kb.
- Includes no
</html>, as they are unrequired.
- Includes as few
"as possible, such as:
<body bgcolor=#ffffff text=#000000 link=#0000cc vlink=#551a8b…
- Includes no comments and almost no whitespace.
- Use smallest tags possible, such as
- Heavily tout use of Google Page Speed to analyze page weight.
- Deliver pages as “chunks”: while back-end is processing, servers can deliver static parts of page that can be delivered prior to server response time.
- Minimize HTTP Requests:
- Sprite everything into a single sprite, even images used infrequently.
- Each new connection increases page-to-load time (latency) by approximately 10%…
- Minify, Obfuscate, and bundle JS files.
- Defer JS downloads for anything unneeded for initial display, using variant of Steve Souder‘s Script DOM Elements, but creating a closed and IDed
<div>located within the
<body>to which all deferred JS is appended (appending to the
<head>can have negative affects in some cases in IE).
- Take advantage of caching images and JS files, working heavily with “chunks” to only refresh the parts of the page that need to refresh.
- Doing a lot of work using <iframe>s and JS “hacks” to work around Ajax History state.
As good as all of the above sounds, and mostly is, I have to say I was a little dismayed when I dug into their code and found that they apparently have no regard for standards, exhibited by them improperly nesting a
<div> inside of a
<div> inside of a
<span id=main><div id=ghead><div id=gbar>), and the use of tags like
<font size=-2>… Tsk tsk…