When performing tests of my site, I’ve observed that my Start Render improves dramatically when I select the Video > Capture Video feature.
This is an unexpected result. I would expect the test results to be unaffected by this test feature.
Here are couple of examples. The first is a test from Los Angeles. First with, and then without, the video selected. And the test after that one compares the same from New Zealand.
Please observe how the selection of the video improves Start Render. It completely changes the relationship between the vertical Green line (Start Render) and the vertical Blue line (Load Time) on the following screenshots.
Thanks, I’ll take a look. I thought I had stomped out the last differences but it looks like there is still something going on (the render and video used to use different methods to grab the screen image but that was fixed a while back).
ok, I have the start render issue sorted out. The non-video path was missing a check at one of the points where video checks which was causing the difference (thanks for helping dig into it). I sucked it up and cleaned up the code pretty substantially based on the work we had done when we implemented the screen captures for Chrome so it was a good win (and there won’t be out-of-order images anywhere anymore either).
I still see a couple of things that I want to look into though:
The Fully loaded time sometimes comes before doc complete. This is because it’s actually the “last activity on the network not counting favicon” time and I think I need to make sure it at least is always at least >= doc complete
Right now I’m being awfully generous to IE by only starting the measurement at the point of the first network activity, not when I tell the browser to start navigating but I’m not extending the same generosity to Chrome. In IE this can easily be more than 0.3 seconds and since we’re trying to optimize pages, not browsers it doesn’t seem fair to include that time (but it does make comparing browsers unfair and in Chrome the time is a LOT faster). I’ll probably see about extending the same generosity to Chrome as we work on polishing up the Chrome support but it’s something to be aware of.
The doc complete line doesn’t always exactly line up with the favicon request. I’m not sure if this is because of the adjustment to the times or if it is just the way it is but I need to look into it more.
Excellent work, Patrick. Thanks for the fix, and thanks for implementing it so promptly.
I ran a few comparison tests, With & W/out video, and needless to say now it works as expected.
I did observe the out-of-order screenshots in the Filmstrip View in the past. It was a bit confusing, but I didn’t mention it because it wasn’t as important (to me at least) as the issue mentioned (and fixed) in this thread. However, since you’ve brought it up, I will mention that I no longer see out of order screenshots, but also I no longer see any screenshots containing background images.
The above link points to a Los Angeles, a test center where my site is fast. But I did test in New Zealand, where my site is slow, and the result was the same (no background image in Filmstrip View).
This background-image issue is for me personally of very minor importance.
Your current fix has greatly improved the testing experience today! Thanks again very much, Patric.
I think the issue you are seeing with the background images is because they seem to get painted after the document complete event (the fully loaded screen shot is actually captured 2 seconds after the network activity stops). If you captured AFT (the above the fold time) it would actually give you a more accurate measurement for when the background images paint (at the cost of having to wait 4 minutes for the test to complete).
I do think that it’s pretty unique to your site because of how freakishly fast it is but it is interesting that the background sprite doesn’t paint before onLoad (is it just a normal image sprite all in css or is any of it manipulated through javascript?).
