Performance/Snappy/2013-03-07

From MozillaWiki
Jump to navigation Jump to search
The printable version is no longer supported and may have rendering errors. Please update your browser bookmarks and please use the default browser print function instead.

Snappy Mar 7, 2013 Minutes

Actions

  • [DONE] benwa to send lmandel list on slow shutdown add-ons for follow-up with add-on team
  • Irving to test what happens when an add-on switches from unpacked to packed (look at ABP specifically)
  • Taras: talk about PICL perf review

Incoming

Projects

Startup & Shutdown

Results
  • bug 845907: Abstracted readahead functions have landed. As a result:
  • bug 810454: A/B test for cookie readahead landed in m-c this morning
  • bug 810151: Ordered jar file readahead - 60% drop in omni.ja startup read time on Windows test. Landed in inbound yesterday, was backed out due to suspected xpcshell orange. Will resolve ASAP.

Tab Strip (tab switch)

Results
  • bug 829952 (slow touchpad scroll): testcase to measure cross-browser + blogged.
  • landed bug 842967 (refresh driver can retarget a timestamp).
  • bug 590422 (Remove timer filter): landed, backed out: lots of backfire from timing-related regressions:
    • bug 845943 (fix talos tscroll) - close to finish.
    • bug 749894 (Mochitest performance.now() orange) -> research -> 10% failures due to disabled high-res timer on XP on 25% of the cases -> 1. test could be improved. 2. regression of bug 822490 (simplify windows timestamp) -> elevated bug 836869 (improve high-res timer detection hueristics - Honza Bambas). Still not decided if low-res timer should pass or fail this test (right now it sometimes fail).
  • (update - mconley): Australis on par with default theme, they moved on to refinements, mconley will occasionally check perf to make sure no regressions.
  • (update - ttaubert) newtab performance: XUL is expensive, consider CSS3.
Todo
  • keep killing regressions of bug 590422.
  • bug 829952 (slow touchpad scroll): talk to masayuki, he has ideas for improvements, see where it stands.

Temporary and Severe Browser Hangs (front-end, downloads, chromehang fixes, scrolling, prefs, slow sql)

Results
  • Final patches for bug 566746 (async form history) posted (including test fixes)
  • bug 820834 landed (about:home storage refactor)
Todo

Add-on and Web page introduced slowness (add-on manager start-up)

Results
  • Found bug 844958, bug 845921
  • Filed bug 846921 for detailed add-on telemetry, got redirected into FHR
  • Discussions with felipe et. al. about next steps for add-on manager improvement
Todo
  • File bugs and finalize design for converting AddonManager.jsm and XPIProvider.jsm from sqlite to JSON
  • vladan to follow up with jorge about slow add-ons

Page Load (cache, dns)

Results
  • Putting up full page load tests to test more than networking layer
  • WIP patches for cache, e-mail try build to taras and vladan

Tools (profiler, eideticker, peptest, talos)

Results
  • SymbolicationServer & Profiler extension to support symbolicating Thunderbird & other apps
  • bug 837724: comparison of Telemetry histograms during a Talos run from a PGO MSVC2010 build versus a non PGO MSVC2012 build
  • Improving security of the profiler add-on for AMO. Parameters are chrome controlled but are passed directly to the shell. I have patches to escape them correctly.
  • Worked with intel to support MSR/power profiling.
  • Discussed with Vlad the profiling needs for Gaming. Planning to focus on Multi-thread/Multi-process support, runtime overhead, JS frame sizes, and delimiting frames and transaction in frames.
  • JS Team is looking into adding profiling support to Baseline & asm.js
Todo
  • Pending feedback: bug 832664 (Efficient, async JSON serialize/deserialize)
  • Mentoring: bug 848278 (Measuring performance of OS.File operations)
  • bug 788495: Reproducing crash in js::SPSProfiler::JMChunkInfo::convert, fix in bug 832812 making it hard although I still get the odd crash once per day