- Still don't know why RSS is so high. 12MB v. IE 2-3 MB
- dprice to look at hot v. cold routines with an eye toward loading less code
- waterson trying to figure out what pages are resident
- Swap in another linux allocator for reduced fragmentation
- kandrot having troubles with what looks like allocator conflicts on SuSE. Redhat Ok.
- Evaluate what's using memory during peak usage on memory hog sites.
- kandrot to look at this
- Examine opportunity for .so and runtime foot print if we can strip exprots and relocations
- cls reports wtc & bryner gave up; only one of the gcc attributes requires glibc support (``protected''); the others just need a modern linker.
- Another thing to follow up on is, can we ``pre-rebase'' Linux .so's to avoid rebasing at load time?
- what is next here? more info on ``magic gcc attributes'' from rth@redhat.com?
- Revisit RTTI bloat issues. gamera is using exceptions which sucks in need for RTTI
- kandrots analysis shows gcc-2.95.2 has 20% higher bloat than egcs-1.1.2
- dont' understand why.
Daily test run and results publishing - twalker/curt
Blackbox Consumption Reports - done
->ftp://ftp.mozilla.org/pub/leak-reports/daily_memtests/memtest_results_daily.html
Develop/port windows memory monitoring tools - mjudge
thesteve to fix tools to rollup 'allocation frequency' by functional area to find biggest fragmentation hogs
Examine the Biggest Consumers of Memory for optimized memory use.
- Image Lib - saari/pnunn - image cache redesign work landing late feb.
- Images are ~20-30% of live object data according to trace malloc reports.
- Largest speed deltas are also on intensive image pages.
- pav, tor - found optimization that allows storage of image bits only on x-server
- evaluting ways to decode to 16 bits on 16bit systems
- empty/spacer images need to be compressed
- Style System - pierre/attanasi -
- look at factoring options
- pierre's latest enabling changes regressed performance. evaluating
- about a 70% space reduction in memory use for sytle rules might be achieveable
- Frame/Frame Manager - buster
- js? (for N6.5?)
- layout?
- i18n/l10n - yokoyama -
- bugs related to footprint and performance (65686,65685,52352,49527,29321,17029)
- Catalog, Control, and Tune Cache and Buffer Sizes System Wide
- . Raw data posted at
ftp://ftp.mozilla.org/pub/bloat-reports/urls-100.log.gz
ftp://ftp.mozilla.org/pub/bloat-reports/urls-450.log.gz
Take-aways
- Gotta figure out why those memcache entries are porkin' up. Can
we do a quick hack while we're waiting for the New New Cache?
Filed http://bugzilla.mozilla.org/show_bug.cgi?id=66692.
- What's all that `void*' stuff? Need to tweak the stack
inference rules.
http://bugzilla.mozilla.org/show_bug.cgi?id=66691 for then HTTPHandler-related growth
- nsStrings? nsVoidArrays? From where?
- 331 new nsHTTPChannel objects. Wow! Why do we have so many
of these lying around? http://bugzilla.mozilla.org/show_bug.cgi?id=66694
image - saari
necko memory - gordon
necko disk cache - gordon
identify others - waterson
tie into new api for memory presure relief - rickg's idea?
Reduce code size with tune-ups to linux compilers
test and migrate
to gcc 2.9.2 - cls, thesteve, and others
optimizations
- ron guillmette
code generation
optimizations - kandrot
explore other
compiler options/optimizations
intel - marek
sony - marek
borland kylix - rickg, porkjockeys
Kill all de leaks, mon...
footprint : 55 bugs found.
mlk
: 90 bugs found- 1/6; 91 bugs found 1/26.