WebKit/Speedometer

Idle time in tests

mram0509 opened this issue · 2 comments

Web browsing and interactivity with web pages inherently includes idle time between actions. How can this be better reflected in the tests included in Speedometer? Further discussion and analysis are required for concrete proposals to reflect the actual experience of users while interacting with a webpage. For instance, idle time could influence CPU scaling and GC, which in turn could influence responsiveness. Would the idle time be customized to a particular test or would it be benchmark wide, that is incorporated in the runner?

rniwa commented

Since we're driving each test case with rAF, there is already plenty of idle time during the benchmark.

bgrins commented

IIRC in the meeting we discussed this in the context of matching longer and realistic delays (e.g. when a user is waiting for some long running async action like a network request to complete, or physical delay between clicks or other input which would be longer than a rAF, etc). It's a tricky problem as we saw in #201, but worth discussing further for v4+, and seems maybe related to supporting async steps (#83).