startup_metric_utils: remove main entry computed ticks
For a little while we have been recording startup metrics based on main entry point time using two techniques: * (old way) by converting process creation wall time using StartupTimeToTimeTicks() * (new way) by calling base::TimeTicks::Now() The new way is theoretically cleaner and removes a few possible skews (see details in the bug). During a few months of overlap we have been computing these two histograms for the identical time interval: * (old way) Startup.BrowserMessageLoopStartTimeFromMainEntry2 * (new way) Startup.BrowserMessageLoopStartTimeFromMainEntry3 On Windows the skew between the two is super tiny (2-3 UMA samples with difference during the day). On Android the difference is between 0.5% and 0.7% for the same day for various quantiles. The overflow bucket (1 hour and more) for the new histogram is 3.5x smaller - nice. Also the new histogram looks a tiny bit smoother on Android. Overall the shape of the distribution does not change our subjective perception of the startup metrics on Android. Based on these observations this change proposes to: * Switch the remaining calculations based on the main entry point to the new way - calling base::TimeTicks::Now() (no conversion to ticks, not using the wallclock) * Avoid renaming the histograms because the expected difference is likely not worth worrying * Celebrate cleanup :) Bug: 797762 Change-Id: I5c32541280c359d6571cbcc5c6e7500b79002011 Reviewed-on: https://chromium-review.googlesource.com/1104696Reviewed-by:François Doray <fdoray@chromium.org> Commit-Queue: Egor Pasko <pasko@chromium.org> Cr-Commit-Position: refs/heads/master@{#568189}
Showing
Please register or sign in to comment