Performance/Snappy/2011-12-15

From MozillaWiki
Jump to: navigation, search

Snappy Dec. 15 Minutes

Action items

  • [2011-12-01] land dom storage - mak reworking it
    • dom storage was blocked on landing SQL timing(mak reworking it)
    • not going to land in 11, working on bugs
  • [2011-12-01] do animation data after profiler is done (Jeff)
  • [2011-12-08] lmandel to follow up with QA about cache efficiency
    • speaking with taras and ahughes
  • [2011-12-08] File bugs on issues in dom
    • Dbaron suggests we can modify our event interface to attribute which webpage is event originating from... avoid attributing to subframe(bug 710927)
    • Tantek suggests having ui that allows us to be able to blame webpages + disable js on that page
    • bug 710935 to measure lag
  • [2011-12-08] Cww to provide more guidance from support for Snappy focus areas
    • (Twitter wasn't useful, sadly).
    • Action for Cww: figure out a way to get back to each twitter user who is complaining with a survey for "why"... probably a Q1-scope task.
    • I'd like to help with looking at telemetry data -- need to focus on the % of users with [worse than X] for each factor.
    • Asa suggested that startup time is a key performance issue

Projects

Front-end Telemetry - Dietrich
Fix cache - tglek, Nick
Profiler - jrmuizel
about
jank - jrmuizel
  • have basic design in my head and hope to have something close to working in the next few days
  • will be an add-on
  • bug #?
  • sounds like bug 703800
ON HOLD - Graphic harness for fps - jrmuizel
  • needs to land for desktop still
Nondestructive chromehang - tglek, bsmedberg
  • bug 706916
  • bsmedberg investigating
  • Won't produce useful stacks without frame pointers?
  • Should we occasionally release builds with frame pointers for a week?
  • [action] Ehsan can talk to releng about this. (new branch?)
  • [action] need to find someone to own this - taras/lmandel
Snappy Scrolling
Interruptible JS engine(luke)
  • Start with leveraging existing infrastructure, may expose bugs.
  • Nested event loops via green threads
  • Starting on Dec 19
Peptest - ahal, mcote
Font chromehang fixes
  • collection of bugs found when turned on chromehang
  • John will aim to have fix reviewed by aurora branch day
  • Given the value of this fix might land it on aurora
Slow SQL Telemetry
  • deployed
  • waiting for metrics frontend
  • need better writeup about when/how to file reports from Slow SQL about:telemetry
  • paste your SQL query in #perf and they'll help you figure out how to file
UX gripes
  • New window takes a long time to open, compare to Chrome or Opera bug 49141
  • After starting the browser, the first door hanger paints in several steps, and the arrow image at the top seems to be slowing it down, maybe it's not loaded up front. Subsequent opening is fast. https://bugzilla.mozilla.org/show_bug.cgi?id=709181
  • New/Close Tab/Tab detach/tab move are slow in terms of perception - dietrich
    • "animations stutter" — try closing a lot of tabs, then moving out of the tab bar and see them resize to fill the tab bar again
    • Frank is working on a direct-manipulation version of move/detach tab (had to be backed out and rewritten): bug 674925
    • Current invalidation code sucks (redraws whole tab bar)
      • Displaylist invalidation will make this better - bug 539356 (no work until new year - Roc)
    • "connecting" in the title of a tab while reloading should be removed, it's visual noise that makes it seem slower. https://bugzilla.mozilla.org/show_bug.cgi?id=709182
    • When opening a new tab, it shows "about:home" for a split second, and (for limi) shows a spinner for a split second)
  • Awesomebar: - dietrich
    • Measure awesomebar drawing with camera - who does this, has this equipment, etc? limi
    • don't show progress spinner while typing — overall, our awesomebar results are now fast enough for this spinner to have lost its purpose — even in the cases where it spins for a few seconds and then returns something slightly different, it's usually too late: bug 702926 (FIXED)
    • URL autocomplete like every other browser: bug 566489 (~1 week away from landing if resourced, according to mak?)
  • Scrolling is worse than opera(acceleration, smoothness, and bounce) - (roc)
    • Turn smooth scrolling off for mousewheel or investigate Bug 206438
    • Turn on smooth scrolling (turn on on nightlies 198964)
    • make sure high res mousewheel scrolling is hooked up on all platforms
    • tweak how far we scroll, physics
    • hook smooth scrolling to refresh driver(smooth scrolling code needs to be updated to not do this separately)
    • track scrolling performance(roc?)
    • Make a video to demonstrate the problem
    • cameron has been fixing touchpad
    • keyboard repeat drives scrolling rate
    • clicking on down arrow
  • Issues with back/forward being slower than opera
    • bfcache isn't used as frequently as in other browsers? - lmandel to follow up with QA
  • Animation in general (progress spinner animation in particular)
    • dbaron suggestions
      • hook up painting to refresh driver
      • track refreshdriver deadlines

Other

  • Performance test day tomorrow - may yield something of value for Snappy
  • Responsiveness criteria for release readiness program
    • think about criteria based on each sub project's work
    • the criteria should be applicable for arewesnappyyet.com
  • Holiday schedule
    • Meeting Dec. 29 is cancelled
  • lmandel will work with engagement, press to publish results from Snappy.