Yeah, confirmed. Main thread looks blocked on the PPAPI process.

How reproducible was it? As in, are we going to be able to see the ad that caused the issue again?

I just reloaded the page again and still see the same issue. So it seems to be pretty reproducible (at least right now).

Also, if you leave the tab open for awhile and then come back, the page gets REALLY janky.

This problem still happens. If you remove the <object> element the jank immediately disappears.

Similar issue it seems on Forbes: CPU spike, fan spinning (even after the Flash video disappears).

Deleted: trace_forbes-fan-spinning.json.gz 13.9 MB

trace_forbes-fan-spinning.json.gz 13.9 MB Download

Deleted: trace_forbes-fanspinning-aftervideoad.json.gz 13.1 MB

trace_forbes-fanspinning-aftervideoad.json.gz 13.1 MB Download

How can we tell who's causing the issue here: Chrome, Flash, Forbes, the ad network?

Adding some additional folks to the issue, who maybe able to advise on how best to diagnose.

It just occurred to me: I have "Plugins" set to the default "Detect and run important plugin content". Shouldn't that have meant that these ads causing the performance problem shouldn't have run? Or are they small one that are exempted?

Looking at the Flash content that's getting loaded (using Flash Control), I have a feeling PPS Tiny might be the correct intervention for this site... this is a pretty good case study in pixel tracking.

--enable-features=BlockSmallPluginContent

Even with that feature enabled, I'm still seeing some content peeking in.

Deleted: Politico.png 477 KB

	Politico.png 477 KB View Download

I've started a deep dive [1], here is what I've found so far:
 1. Hammered CPU: several third parties doing viewability without IntersectionObserver [2]
 2. Hammered CPU: some over-triggering third party code
 3. Scroll jankiness: at least one third party is using a wheel event handler, the generic recommendation is to consider Passive Event Listeners. However, the third party use case being viewability, the better recommendation is IntersectionObserver. 

IntersectionObserver is the right answer to 1. and 3. I'm getting in touch with the relevant stakeholders.

For 2:
 a. the upcoming intervention to throttle timers of offscreen frames might help provided that the culprits are in sub frames ( issue 616519 ).
 b. regardless of the intervention noted above, this doesn't feel like a sane behavior so I've reached out to the owners of that code.




[1]: https://docs.google.com/presentation/d/1172haFf9Aw8Abpyxgx9cpTs8cXQFBvP_GCjXRDruv4Y/edit#slide=id.p

[2]: http://blog.chromium.org/2016/05/new-apis-to-help-developers-improve.html

Forbes and Politico are also impacted by  issue 626202  (triggered by an increasingly janky scroll on AndroidAuthority), in particular the findings in comment #14 to #17.

Probably not much additional insights, closing.