lanwei@ gathered data from 2,319 deep reports. These are traces from employees during regular use, sampling for the first 30 seconds after each navigation.
We examined the time spent when touch start and touch move events were queued on the main thread, or being handled on the main thread.
By “queueing time”, we’re referring to the time when the event has reached the main thread, but the main thread is busy doing other work.
By “handling time”, we’re referring to the time when event handlers are actively executing.
This chart shows the average contributions for the top 20 methods by contribution to latency. The “worst” rows show the average contributions for the worst 10% of cases.
This chart shows the same data as above, but normalized to make it easier to compare the proportion of time spent in each method. It’s interesting that the proportion of time spent per method is extremely similar in the “worst” cases and the cases including all data.