View source for Performance/Snappy/2012-02-16

=Snappy Feb. 16 Minutes=
==Actions Items==
*Tab on demand decision - gavin
** https://bugzilla.mozilla.org/show_bug.cgi?id=711193#c34
** more work to measure session restore responsiveness - should have data Friday or Monday
* QA should verify tab close regression in bug 724349

==Incoming==
* 20 second GC on resume (bug 639262) - dietrich

==Projects==
;Better DOM event/task scheduling - jst(telemetry)
* {{bug|712478}}: Add interactive states(needs owner):
** user-interacting: using is actively scrolling, clicking, entering fullscreen({{bug|715592}}), etc
** user-might-interact: ie mouse moved recently, might be followed by a click, do not schedule low priority timers within 50ms
* {{bug|715376}} & {{bug|715378}}:  background tab activity to be less disruptive
* {{bug|715380}} to measure above problem + track pro gress (nathan)
** {{bug|715950}}
** {{bug|715953}} checked in
* {{bug|717659}} for tracking reflows in background tab
** marked checkin-needed
* Avoiding reflowing when scrolling ({{bug|715584}})
* throttle parsing tasks while loading background tabs(html5 parser work bug#)
* timer combination  + time accounting on JS are being assigned(eta Thurs 23rd for update)

;Interactivity heuristics - taras, mccr8, bbondy 
* reduce frequency of cycle collector when slow ({{bug|710496}})
* try to disable cycle collection, gc, while user-interactive ({{bug|698547}})
* mccr8 will look into {{bug|710496}} by next meeting

;Super-slow-startup investigations - vlad, taras, dietrich
* Main bug https://bugzilla.mozilla.org/show_bug.cgi?id=701872
* network cache still a problem?
* a lot of slow startups seem to indicate we load pages before we paint: {{bug|715402}} (backed out, need to re-evaluate Talos tests w/ new startup event sequence caused by this patch)
* Vlad's blog post: http://blog.mozilla.com/vdjeric/2012/01/31/telemetry-from-super-slow-startups-first-impressions/
*Unknown what % of slow-startup reports were from unofficial builds, getting recent batch of data with extra info in METRICS-429 JIRA
* Vlad looking into slow sql correlations: reported findings to mak during FOSDEM week
** Co-ordinate with froydnj, maybe have SQLite add callback on lock contention

;Startup optimizations - bbondy
* Misc:
** Updated my local profiler to only time things on the main thread.
** Wonder if we can make a smaller dll instead of xul.dll with the needed startup functions
** Looked into the preload/LoadLibrary code
* In Progress:
** Needs review comments implemented, {{bug|725444}} - 10-15ms in AudioSession
** Other opts for nsLocalFileWin, would also like to profile nsLocalFileUnix later
** Profiling manifest parsing
** Searching for other startup pain points
** Investigate delayed loading of dynamic libraries
** Tweak preload code for when prefetch is disabled
* In Review:
** {{Bug|692255}} - Find a way to get rid of prefetch files on Windows for faster startup
** {{Bug|727864}} - Enable DLL preloading when MozillaMaintenance service disabled prefetch
* Landed, or landing soon:
** {{Bug|726503}} - FILE_FLAG_SEQUENTIAL_SCAN always being passed for all file IO.
** {{Bug 726481}} - Reduce calls to IsDirectory by not calling it when enumerating directories.
** {{bug|726576}} - Optimize calls to nsLocalFileWin's HasFileAttribute

;Front-end - Dietrich/bbondy
* {{bug|671038}}: front-end telemetry meta bug
** recently landed: number and size of Places annotations, frecency updates, opening of the app menu, opening of bookmarks toolbar, time for idle maintenance, idle frecency update, page thumbnail rendering and storing.
** session restore probes just need landing
* {{bug|699820}}: Storage IO meta bug
** recently landed: {{Bug|722242}} - (placesIdleContention) Avoid thread contention on idle maintenance
* {{bug|669034}}: Session Restore meta bug
** breaking out chunks into separate modules, some patches in review
* {{bug|593680}}: Tab animation meta bug
* {{bug|711193}}: Restore-on-demand by default (needs telemetry, see top of page)
* {{bug|704025}}: Rewrite frecency to make it less IO heavy (needs owner)
* {{bug|725821}}: Avoid scanning all the pages table for each awesomebar  matching,  separate FTS engine to restrict the resultset (needs owner)
* {{bug|519514}}: Async bookmarking API to avoid Sync hanging the UI (needs owner)
* {{bug|699860}}: Async Clearing of recent history (needs owner)
* {{bug|566746}}: Async FormAutocomplete in reviews (needs new patch, emailed felix to see if he can still work on it)
* {{bug|699856}}: Move search service to json (yoric, needs review gavin)
* {{bug|699859}}: Async content preferences (drew, needs new patch)
* {{bug|521264}}: Don't use file.exists() when not necessary (marcoC, failing tests, needs new patch)
* {{bug|649216}}: Clicking quickly at the same location should close multiple tabs (previous owner may be able to come back to this - follow up)
* {{Bug|710373}} - Pressing the vertical slider down arrow with smooth scrolling should move three lines instead of one (Jared)
* {{bug|613588}}: asyncronous on-demand livemarks (marco, needs review dietirchy)
* bbondy: {{bug|727275}} - 1s opt main thread for a download from Download Manager (implementing review comments)
* bbondy: {{bug|632556}} - Run nsIFile::Reveal/Launch asynchronously (waiting on feedback from bsmedberg)

;Fix cache - hurley
* work continues, see https://bugzilla.mozilla.org/show_bug.cgi?id=717761
* https://bugzilla.mozilla.org/show_bug.cgi?id=715752
* https://bugzilla.mozilla.org/show_bug.cgi?id=105843
* cache testing http://groups.google.com/group/mozilla.dev.tech.network/browse_thread/thread/6dfa3e6ebe80c336
 https://docs.google.com/spreadsheet/ccc?key=0Amywl0iz5SaydEhwblpOQng1RkpoS09Yakh6NkxoRnc 
** Not really BFCache, but some results of Click-through Caching here -> https://docs.google.com/spreadsheet/ccc?key=0AmkRt0ylPb8zdGxPRDhUbGdoM3lwd3JQNXY3bW5nNHc&hl=en_US#gid=0

;Profiler - jrmuizel
* Have plans to get samples associated with document urls
* sfink is investigating interleaving C++/JS stack in profiling unwind
** BenWa: Filed {{bug|719530}}
* No update - on hold for a few weeks

;about:jank - jrmuizel
* changing the profiler export format to JS
* No update - on hold for a few weeks

;Paint telemetry - jrmuizel
* No update - on hold for a few weeks

;Nondestructive chromehang - vlad
* First draft complete https://bugzilla.mozilla.org/show_bug.cgi?id=712109 (patch not landed yet)
* Getting PCs from stack + module addresses & PDB signatures
* Client-side reporting of chrome hang stacks + module PDB signatures done. Offline symbolication of stacks works. Latest patch with leaner report format currently being reviewed
* Benoit has made a symbolication web app, might integrate my symbolication tool with it eventually
** I've taken over getting Beonit's symbolication web app deployed, currently working on getting access rights from Server Ops

;Snappy Scrolling - Jared 
* (2/15 update: A patch to make keyboard arrow keys and scrollbars have consistent scrolling distances is very close to landing. The patch will introduce a pref that can be changed in about:config for scrolling distance. Hooking smooth scrolling up to the refresh driver timer is no longer part of the snappy scrolling work since it will not have a large impact on the responsiveness of scrolling. The last major work for snappy scrolling is {{bug|206438}} which needs an owner. Jared has reached out to the extension author to see how he can help us with porting it to the platform.)
* https://bugzilla.mozilla.org/show_bug.cgi?id=710372
* (add backend dependencies)
* Scrolling UX Gripes
** Scrolling is worse than opera(acceleration, smoothness, and bounce) - (roc)
*** Turn smooth scrolling off for mousewheel or investigate
*** Smoothscroll should be replaced by 'smoothwheel' extension - Bug 206438 (needs an owner, Jared reached out to smoothscroll extension author on 2/15)
*** make sure high res mousewheel scrolling is hooked up on all platforms
*** tweak how far we scroll, physics
*** track scrolling performance(roc?)
*** Make a video to demonstrate the problem
*** cameron has been fixing touchpad
*** keyboard repeat drives scrolling rate

;Peptest - mcote
* Peptest back on try any day now...
* {{bug|719618}}: still waiting on one more patch to land
* {{bug|719511}}: mozharness fixed, should be re-enabled todayish.

;Font chromehang fixes - John
* collection of bugs found when turned on chromehang
* John will aim to have fix reviewed by aurora branch day - complete?
* Given the value of this fix might land it on aurora - status?
* Jeff will follow up with John to try and get some help for this issue
* John: updating patches for system font fallback today, 1/27

;GC pause reduction - billm
* incremental GC ({{bug|641025}}), responsiveness telemetry ({{bug|719492}})

;CC pause reduction - smaug, mccr8 (meta {{bug|698919}})
* landed more patches to remove objects from CC graph (meta bug: 716598)
** object holders ({{bug|725768}}), content list ({{bug|725446}}), XBL protos ({{bug|725752}}), anon content ({{bug|725867}})
* run the CC less often by pruning purple buffer (meta {{bug|726442}})
** landed {{bug|726331}}, {{bug|727572}}, {{bug|727625}}
** {{bug|726331}} stops CC from running on TechCrunch while idle (see {{bug|725377}})
* pause time telemetry for new CC phase ({{bug|726007}})
* more compartmental GCs ({{bug|716014}}) waiting on incremental GC

;gfx d2d lag - jeff
* need better heuristics to turn it off
* need a plan to make us paint as well as IE

;Firefox shutdown - rafael/bbondy
* {{bug|662444}}
* shutdown quickly and properly
* affects places, session restore, etc.
* bbondy: A large portion of shutdown is due to the main thread's  nsDiskCacheBlockFile::Close/nsDiskCacheBlockFile::FlushBitMap accounting for 193ms
* bbondy: {{Bug|685027}} - Closing last Firefox window takes longer to exit the process from Firefox > Exit (Was nothing was just due to notification service), will follup up with another bug but it isn't highest priority

;ImageSuck
* first step is {{bug|689623}}, tnikkel is working on it now