Performance/Snappy

Snappy is a project that aims to improve Firefox responsiveness. This project is a responsiveness equivalent of Performance/MemShrink.

Meetings

Meetings will happen every Thursday at 11am Pacific time.

• Dial-in: Audio-only conference# 95346
  • People with Mozilla phones or softphones please dial x4000 Conf# 95346
  • US/Toll-free: +1 800 707 2533, (pin 4000) Conf# 95346
  • US/California/Mountain View: +1 650 903 0800, x4000 Conf# 95346
  • US/California/San Francisco: +1 415 762 5700, x4000 Conf# 95346
  • US/Oregon/Portland: +1 971 544 8000, x4000 Conf# 95346
  • CA/British Columbia/Vancouver: +1 778 785 1540, x4000 Conf# 95346
  • CA/Ontario/Toronto: +1 416 848 3114, x4000 Conf# 95346
  • UK/London: +44 (0)207 855 3000, x4000 Conf# 95346
  • FR/Paris: +33 1 84 88 37 37, x4000 Conf# 95346
  • Gmail Chat (requires Flash and the Google Talk plugin): paste +1 650 903 0800 into the Gmail Chat box that doesn't look like it accepts phone numbers
  • SkypeOut is free if you use the 800 number
• Vidyo: PB&J
• IRC: #perf
• Etherpad: snappy (copied to wiki after the meeting)

Minutes and Progress Reports

• Minutes Thu, Dec 1, 2011
• Minutes Fri, Nov 18, 2011

Bug Tracking

Bugs tracked by the Snappy project are prioritized by adding one of [Snappy:P1], [Snappy:P2], or [Snappy:P3] to the whiteboard. Use whiteboard [Snappy] tag to nominate bugs.

• Unprioritized Snappy bugs. These are triaged regularly in meetings.
• Snappy:P1 bugs. These are discussed regularly in meetings.
• Snappy:P2 bugs. These are discussed occasionally in meetings. P2 is the default priority.
• Snappy:P3 bugs. These are discussed rarely in meetings.

Some other interesting bugs/lists.

• visible unresponsiveness metabug
• sync storage main-thread-io metabug
• cycle collector responsiveness metabug

Goals

• 50ms responsiveness when typing in a textbox (bug 703668)
• 60fps animations on UI operations, i.e. closing tabs (bug 702509)
• track above via telemetry

more goal thinking responsiveness goals

My (Asa) proposed core measures:

• Startup on par with IE and Chrome
• 50ms responsiveness for all non-animated primary UI interactions (list TBD)
• "Above the fold" uncached page load on par with IE and Chrome
• Back/Forward navigation as fast as Opera.

Current Infrastructure

• Event Tracing: http://mxr.mozilla.org/mozilla-central/source/toolkit/xre/EventTracer.cpp#38
  • Fires events at main thread event loop from background thread, measures response time
  • Produces output of the form MOZ_EVENT_TRACE sample <timestamp> <duration>
  • Being rolled out as part of Talos Tp5 bug 631571, will produce a responsiveness metric for "responsiveness while running Tp5"
• Peptest: bug 674606
  • Responsiveness regression test harness
  • Allows devs to write tests that perform various actions and tests responsiveness using EventTracer
  • Project page contains everything you need to know
  • Run 'make peptest' from your object directory to run the tests (see running tests for command line instructions)

• Telemetry dump for ff11 from November 17

• Talos responsiveness tests - for details see bug 631571

Additional Infrastructure

• Have metrics do continuous trend analysis to see if our telemetry stats are getting better over time.

arewesnappyyet.com

• This should use our telemetry data to tell us whether we are meeting our expectations and how how far off we are
• bug 703669 tracks the implementation of this site

Testing

Placeholder section for QA and testing.

Sprint 1 Phase 1

Takeaways from 2011-12-01 Meeting

• Goal of Project:
  • Big hangs (~30s) are a problem, but small hangs (~5s) may be a bigger win
  • Identify where we hang, why we hang, and plug those holes
• Things on Devs Radar:
  • Meta Bugs: visible responsiveness (bug 698500), sync storage (bug 699820), cycle collector (bug 698919)
  • Animation hangs are most visible
  • Windows, Mac, Linux are all unique -- Windows is P1 for the project
  • Firefox Built-in Profiler: late Q1'2012
  • Chromhang Bugs: bug 705761, bug 705287, bug 705258, bug 705594, bug 704933, bug 699051
  • Small Jank: using refresh driver to measure refresh-rate performance (ie. throbber, progress bar, etc -- Asa has more ideas)
  • Measuring Snappiness: Telemetry use cases
  • Metrics/Automation: for trend-analysis, finding regressions with Telemetry
• Where does QA fit in?
  • P1: Desktop responsiveness competitive testing
  • many of these problems need stack traces (something Telemetry cannot do)
  • crowd-sourcing use cases where we are less Snappy
  • A-B testing with other browsers (versions of Firefox, competitor browsers)
  • Ownership of Snappy P1s
  • Caching issues should have patches landed in the next week or so