Just a heads up, and the post notes this, that this way of timing is already well into the page load sequence: the server has received the request, built the HTML and has started sending it, and the browser is already parsing it.<p>Also, be careful with when different browsers fire the onLoad event. For example, Safari 3:
<a href="http://www.howtocreate.co.uk/safaribenchmarks.html" rel="nofollow">http://www.howtocreate.co.uk/safaribenchmarks.html</a><p>Overall though, a cool hack even with the caveats above as it could give you actionable data.<p>Another good source of loading time metrics is Google Webmaster Tools. They have two metrics, one in the download time in the Crawl Stats section the Site Performance metric in the Labs section. Combined, they give you a really good insight into how your website is doing over time. A good game to play is spot the (very clear) inverse relationship between the number of pages crawled vs download time.
If after timing your page you want to get some external confirmation I use this site frequently:<p><a href="http://www.websiteoptimization.com/services/analyze/" rel="nofollow">http://www.websiteoptimization.com/services/analyze/</a><p>It gives some pretty useful results when you're tuning your page because it breaks it down in to components.
Clever idea!<p>Although I prefer Chrome Dev Tools for this task - measuring page load time of people with different connection speeds, CPUs (for rendering), browsers seems a little weird.<p>The graph about bounce rate caught my attention. Although the author draws a straight line, it looks more like a parabola with minimum around 3.5 seconds. Which is kind of unexpected (and I can't even think of any reasonable explanation for this).
This is a nice idea, so points for ingenuity.<p>There are a few small problems though:<p>- Page 'load' time is a bit ambiguous, I'd prefer page 'rendering' time, perhaps. (The author does note the limitation of this approach.)<p>- Time to window.onload may be a _very_ long time in some environments, especially when 3rd party scripts are present, especially (if memory serves) in IE.<p>- From eyeballing the "Bounce rate by page load time" the linear trend placed on top seems dubious -- if anything bounce rate seems to drop until the 3700ms mark. (The author notes in the text however that users seem tolerant up to the 5000ms mark).<p>Quibbles aside, having more page rendering time data correlated to bounce rate is an excellent idea, as I've seen some posts in web design land get pretty carried away with n-th degree optimization because they read speed matters for Google and Amazon.<p>In my experience, I optimized a site's page rendering time dramatically, expecting a bump in average pages per visit, and saw absolutely no change whatsoever. We need more data :)
If you're interested in more aspects of web page performance, Show Slow (<a href="http://www.showslow.org/" rel="nofollow">http://www.showslow.org/</a>) is a good tool.
Like the result from the article. Showing the page load time vs percent of users is a very useful way to analyze site performance:how many percentage of site users are experiencing slow performance?<p>Further, combing this with web timing API would be even more interesting.