Closed Bug 1198650 Opened 10 years ago Closed 10 years ago

[e10s] Develop a jank per minute of active usage performance metric

Categories

(Toolkit :: Telemetry, defect)

Product:
Toolkit
Component:
Telemetry
Type:
defect
Priority:
Not set
Severity:
normal

Tracking

()

Status:
RESOLVED INCOMPLETE
Tracking Flags:
Tracking Status
e10s + ---
firefox43 --- affected

People

(Reporter: jimm, Unassigned)

References

Details

User Story

Currently we are using very rough formulas for quantifying a session's performance. A better formula would allow more meaningful comparisons of e10s vs non-e10s performance.

As inputs to this formula, we need:

1) Better jank measurements
2) Measurements of scrolling quality & page-load delay
3) Better measurements of user active time
4) User feedback (e.g. via Heartbeat surveys) on how they perceive Firefox's performance during a session
Develop a "jank per minute of active usage" metric. There is some blocking work associated with this, plus we'll need some sort of HUD/script for extracting this information in a timely way for nightly/release builds.
Depends on: 1107782
User Story: (updated)
Depends on: 1223895, 1198199, 1198196, 1221674
Depends on: 1223897
Sounds like we can reuse data from the vsync driver.
Assignee: nobody → dteller
I may be missing something, but how is this different from FX_REFRESH_DRIVER_CHROME_FRAME_DELAY_MS/FX_REFRESH_DRIVER_CONTENT_FRAME_DELAY_MS?
Flags: needinfo?(jmathies)
Sounds usable, although the user story details something more robust than a single measurement. This came out of Vladan's rc requirements document, ni'ing him.
Flags: needinfo?(jmathies) → needinfo?(vladan.bugzilla)
Yes, the user story has several inputs, but afaict, this bug is only about input 1.
> 1) Better jank measurements This work is already underway in other bugs, it's the work to verify BHR, EVENTLOOP_UI_LAG_EXP_MS and FX_REFRESH_DRIVER_* probes > 2) Measurements of scrolling quality & page-load delay Scrolling quality for the non-APZ case is covered by FX_REFRESH_DRIVER_SYNC_SCROLL_FRAME_DELAY_MS. The APZ case is not covered. It turns out we don't have good Telemetry measures of page-load delay, so there's outstanding work there. See https://bugzilla.mozilla.org/show_bug.cgi?id=1236520#c7 > 3) Better measurements of user active time Chris Hutten has already fixed user activity detection. > 4) User feedback (e.g. via Heartbeat surveys) on how they perceive Firefox's performance during a session This is bug 1223895. ---- This bug is about deciding how to use the numbers obtained from probes #1 and #3 in our server-side analyses. So far we've been using "rate of BHR hangs > 125m per minute of browser uptime" as *THE* jank measurement metric. There might be better ways to analyze the data.
Flags: needinfo?(vladan.bugzilla)
Assignee: dteller → nobody
Benjamin, is something we need to keep around?
Flags: needinfo?(benjamin)
No longer depends on: 1223895
No, I don't think this tracks anything we're going to prioritize.
Flags: needinfo?(benjamin)
Status: NEW → RESOLVED
Closed: 10 years ago
Resolution: --- → INCOMPLETE
You need to log in before you can comment on or make changes to this bug.