The goal of this finch experiment is to estimate the performance and stability impact of running V8s --runtime-call-stats flag which is used in all our performance tests to gather raw internal metrics.


The ultimate goal is to possibly collect runtime call stats (RCS) for a small population of users to get a detailed understanding of how JavaScript is used in web.

In order to evaluate whether this makes sense or not we have to measure the performance and stability impact of RCS. Given that our real world benchmarks run with RCS enabled we don't expect much impact on the stability side. However we do expect a 10 to 15% performance regression with RCS enabled.

On the metric-side we're mostly interested on the impact on the V8 metrics (V8.Compile, V8.Execute) and probably some startup metrics.

FYI, for the "ultimate goal" (collecting new data from users), that would likely require a launch bug and commensurate privacy review.  (It depends on what you're collecting.)  Simply seeing whether enabling this flag slows down Chrome is fine for this as a non-launch bug.

SGTM. Given that we will collect more detailed information on where you spend time in the VM we have to think about potential privacy issues.