I believe we can break out media. "initial page load" is non-trivial to define. What are your goals w.r.t. initial page load bytes? I'm disinclined to use any non-user-visible moments like the onload event (or domcontentloaded). We're working hard to move away from onload so I'd like to avoid coupling new metrics to it if possible.

Time to first interactive isn't implemented yet, but I think it would be a reasonable choice here.

If we wanted something here before that's done, we could maybe use some rough heuristic like "all resource requests initiated before first contentful paint", or something along those lines.

I think splitting out by file type would be less heuristic driven, and potentially more useful. Splitting out JS, CSS, and HTML would be quite interesting.

TTI seems like a great choice, once it's available. Splitting by file type would be informative as well.

This would need a UKM review and a more concrete definition of how this data is broken up.

Refreshed during triage.

I can confirm that Time to Interactive is available now. We have InteractiveDetector[1] on the renderer, and the timestamp is plumbed to the browser which can be accessed as a page load metrics observer[2]

Unfortunately, due to the nature of this metric, we can only record it successfully around 30% of the time right now [3]. If you do use it, you will have to be careful about this. 


[1] https://cs.chromium.org/chromium/src/third_party/WebKit/Source/core/loader/InteractiveDetector.h?l=28&rcl=2f43d0a5b554eb1eb41c28f1a5d5c57c955710a8
[2] https://cs.chromium.org/chromium/src/third_party/WebKit/Source/core/loader/InteractiveDetector.h?l=28&rcl=2f43d0a5b554eb1eb41c28f1a5d5c57c955710a8
[3] https://uma.googleplex.com/p/chrome/histograms/?endDate=20180305&dayCount=7&histograms=PageLoad.Experimental.NavigationToInteractive%2CPageLoad.Experimental.TimeToInteractiveStatus&fixupData=true&showMax=true&filters=channel%2Ceq%2C4%2Cisofficial%2Ceq%2CTrue&implicitFilters=isofficial

Refreshed during triage.

There is a seperate bug for video bytes, and I'd like to keep this bug scoped to just bytes before/after a certain event.

https://bugs.chromium.org/p/chromium/issues/detail?id=821998

I'd propose recording bytes used between the following events:

Start | FirstContentfulPaint | TimeToInteractive | End

Adding FMP to the chain would be good too, if we believe in that metric.

Re: do we believe in FMP, in my experience it works fairly well ~80% of the time, and fails pretty badly in the remaining 20%.

In aggregate, the failures tend to average out fairly reasonably.