Example trace from N5X: https://console.developers.google.com/m/cloudstorage/b/chrome-telemetry-output/o/http___bbc_co_uk_2018-02-05_17-38-56_31963.html

After the main resource is fetched, there is ~200ms of network idle time due to waiting for the "HTMLDocumentParser::processTokenizedChunkFromBackgroundParser" event to kick in.

We have two events "ChildProcessConnection.start", 1st one for the GPU process, 2nd for the renderer. (The first one responds with onServiceConnected in 150-160ms, the second one is faster- 50ms .. this difference is probably due to the UI thread being jammed).

The renderer child process connection is suspiciously close to profile initialization, which is explainable, though it will probably be more optimal to issue the two connections right after the native library is loaded (or loaded+prefetched - not sure).

This optimization can help with FirstContentfulPaint, which we do not record yet on the new startup benchmark.