Performance/Snappy/2012-08-02

From MozillaWiki
Jump to: navigation, search

Snappy Aug 2 Status (no meeting this week)

Actions

  • taras to speak with jst about measuring compartment overhead

Incoming

  • bug 778718 startup regression
  • bug 777266 we should block obviously crap addons
  • Q3 goal for Frontend

Projects

Various frontend bugs (ttaubert)

Results from the past week
  • worked on bug 728294 - Analyze cycle collection logs after test runs to detect leaks
Todo this week
  • finish bug 728294 - Analyze cycle collection logs after test runs to detect leaks
  • get bug 650968 (Enabling a Persona causes significant startup slowness) reviewed
  • get bug 753448 (preload newtab pages in the background) reviewed
  • get bug 779618 (make tabbrowser use the thumbnail service) reviewed

Tab switch performance - jrmuizel

Results from the past week

Better DOM event/task scheduling - nathan (telemetry)

Results from the past week
Todo this week
  • Respond to review feedback
  • Continue chasing down remaining test failures
  • Measure performance impact

Front-end - Gavin, Jared and Frank (Australis tab strip)

Results from the past week
Todo this week
  • paolo is going to investigate bug 699839 (sync IO in XPIProvider.jsm)

Fix cache - hurley, bbondy

Results from the past week
  • Finished this over the weekend: bug 777328 - Gather telemetry data for how much cache corruption reduction plan would help. Waiting on review. It is basically 1/2 of the actual work we'll need to do for reducing the cache corruption in bug 105843. It will also tell us by how much we can decrease the 10% cache corruption in Windows/OSX and 20% cache corruption in Linux.
Todo this week
  • After getting a review on bug 777328 I'll implement review comments until it gets landed, hopefully this week.

GC pause reduction - billm, jcoppeard

Results from the past week
  • bug 777919 - Free LifoAlloc chunks on background thread, instead of as part of the final IGC slice. This isn't a problem for most people, but for some people on OSX it can take anywhere from 50ms to 250ms or more.
  • bug 767209 - Make GC slices longer when not painting to avoid non-incremental GCs.
  • bug 778993 - Separate runtime's gcMallocBytes from compartment's gcMallocBytes, so we trigger less non-incremental GCs with many tabs open.

Startup optimizations - bbondy

Results from the past week
  • Finished reverting the prefetch deleting stuff
  • Now we always do preload on Windows because Telemetry showed that to be optimal, we are no longer gated on previous heuristics on IO activity.