Web Performance: What is the best way to determine "real" page load time?

Speed IndexPat Meenan, the creator of WebPageTest, invented a new metric in 2012 called Speed Index exactly because of the problem you described. The onLoad event just isn't representative of the way pages load and when users feel the page is actually done loading.Here's how it works. When you run a page through WebPageTest, it is continuously taking screenshots of the page while it loads. When the content stops appearing on the screen, it is determined to be 100% visually complete. Conversely, the blank white page it started out as is considered 0% complete. As areas of the page progressively appear on screen, the page becomes some percentage more complete. If content disappears during the page load, of course this would decrease the completeness.Compare the visual progress of these two pages: The page on top loads most of its content (93% as noted below the screenshot) in one second. The other one loads much less content (only 18%) in the same amount of time. Plotting these percentages over time, we get this chart: The page that loads more content sooner "feels" like it loads faster. The problem is, it takes 12 seconds for both pages to reach total visual completeness. The onLoad event is a sucker for waiting for the last image to be rendered and it doesn't care if you could have started reading the page 10 seconds ago. So we use this notion of visual completeness and a bit of calculus to come up with a Speed Index value. The "faster" page has a Speed Index of 1219 and the slower page has a more expensive value of 9022. Just like page load times, higher values are slower and worse. The values are determined by taking the area of the chart above the progress line. The formula for this is [math]Speed Index =\int_0^{end}(1- \frac{VC}{100})[/math] where [math]end[/math] is the end time in milliseconds and [math]VC[/math] is the percent of visual completeness.See https://sites.google.com/a/webpagetest.org/docs/using-webpagetest/metrics/speed-index for more information.Custom MetricsVisual completeness is not always an accurate measurement of load time, because "loaded" is such a subjective concept. A news page is considered "loaded" under totally different circumstances than a video page. I work at YouTube, for example, and we consider our video pages to be loaded when the video starts to play. What is the browser-level event for when a video plays? There is none. A video is continuously changing the picture on the screen, so when is it really visually complete? That's a tough question that Speed Index doesn't necessarily account for.The solution is to use custom metrics. If the application knows that the video has started playing, it can register that moment in time using the http://www.w3.org/TR/user-timing/. For example:performance.mark('playback_start');This would signify the completion of the page load. Tools like WebPageTest can support custom metrics like this in two ways. First, it includes custom metrics in the test report. See http://www.webpagetest.org/result/151015_2C_72H/1/details/ for these test results.Above where it says "Waterfall View" note the column "test" which is the label of a user timing mark. In the waterfall chart itself, this mark is shown as a purple line with a solid arrow pointing down at it. You can see this line on the left edge of the pink area.The second way in which WebPageTest supports custom metrics is with Client Side Instrumentation (CSI) beacon parsing. I won't really go into that because it's mostly a Google-specific thing, but consider this request on a Google search result page: http://www.webpagetest.org/result/151015_5G_6HS/1/details/#request7*. Note the parameters biw=1014&bih=650. This is the search results page itself beaconing its custom metrics to the Google server for analysis. WebPageTest has the capability to parse out the biw/bih values and allow you to use it like any other metric in the test analysis.See http://www.html5rocks.com/en/tutorials/webperformance/usertiming/ for more information.* the test results have since been deleted, but you would have seen a beacon including some performance data transmitted to a server for aggregation.

If you know which elements of the webpage you need in order to declare it loaded, then you can just look at load time of those elements. For example if most of your webpage is visible when you have the HTML and .css loaded, then you can look in the browser / Developper Tools / Network tab and you can get your response time from there.If your question is what tools you can use to get this information automatically, then that is a great question, please let me know if you find out the answer for that.