| Peter | |
| Aug 29 2016, 6:05 PM |
| F4433727: Screen Shot 2016-09-05 at 3.14.22 PM.png | |
| Sep 5 2016, 1:15 PM |
| F4413902: Screen Shot 2016-08-29 at 8.02.09 PM.png | |
| Aug 29 2016, 6:05 PM |
Before we switched to HTTP2 and running on the small(er) agent instance we had some increase in Firefox that happen for a couple of days and then stopped.
Looking at the current graph it seems that we now have the same scenario + that we have had increases before (12/8 and < 1/8).
Lets check the screenshots and waterfall graphs to check if there's something we can make out of it.
This is good example to investigate if we can understand why we have the difference:
http://wpt.wmftest.org/video/compare.php?tests=160830_ET_2M%2C160822_3W_64&thumbSize=200&ival=100&end=visual
Couldn't find something in the waterfall (need to check more though). The domInteractive has moved, it happened between 2016-08-25 13:32 -> 17:24.
On the navigation time per browser, it's hard to actually see something for Firefox:
https://grafana.wikimedia.org/dashboard/db/navigation-timing-by-browser
Lets test out the debugger in Firefox and see I can see something.
This went away for a couple of days and then came back.
This compare WebPageTest runs 3 vs 5 September.
The things Firefox is doing better now. What's really interesting is that the async cache-eval load-time that generates the regression in our navigation timings 95/99 percentiles gives us better SpeedIndex for Firefox. Without the nav timings we had missed those regressions.
Closing this now since it looks ok and I haven't seen those increases last month.