Open Bug 1851924 Opened 2 years ago Updated 2 years ago

Time spent in NtAlertThreadByThreadId in TaskController in speedometer 3

Categories

(Core :: XPCOM, enhancement)

enhancement

Tracking

()

ASSIGNED

People

(Reporter: alexical, Assigned: alexical)

References

Details

(Whiteboard: [sp3])

I recently did an experiment where I sidestepped the TaskController stuff for Ion-related tasks and instead used a pair of dedicated threads and the simplest scheduling mechanism I could write that still slept and woke the threads as usual. It yielded a 1% speedometer 3 improvement: https://treeherder.mozilla.org/perfherder/compare?originalProject=try&originalRevision=bf6b80d429bde27f855f76b9046a1c1492487e87&newProject=try&newRevision=a970576a259395c8eb4bf7026eef7d7f9efff708&framework=13&page=1

Here is a profile without my change: https://share.firefox.dev/3Z72Nul
And with my change: https://share.firefox.dev/3Er4MQX

I'm not sure why this is not incurring the same overthread for waking the necessary thread, but I'm going to be looking deeper at this. I suspect there isn't a barrier to realising this win in the TaskController itself, but we'll see.

Whiteboard: [sp3]

IIUC nothing is really broken for our users, it is "just" a potential enhancement.

Type: defect → enhancement
You need to log in before you can comment on or make changes to this bug.