Closed Bug 631705 Opened 13 years ago Closed 10 years ago

profiling membuster

Categories

(Core :: General, defect)

x86
Linux
defect
Not set
normal

Tracking

()

RESOLVED INCOMPLETE

People

(Reporter: sayrer, Unassigned)

References

Details

(Whiteboard: [MemShrink:P3])

Attachments

(1 file)

an instrumented copy of membuster lives here:
http://people.mozilla.com/~sayrer/2011/membuster.html

http://people.mozilla.com/~sayrer/2011/02/membuster-stats/linux3.6.html
http://people.mozilla.com/~sayrer/2011/02/membuster-stats/linux-mc.html

These stats show that m-c is carrying 50-100 DOMWindows more than 3.6, and m-c leaves 200 DOMWindows around after the run has completed. The detailed m-c stats show that image-data, mjit code, and layout (to a lesser extent) are the bigger bits. js/gc-heap doesn't shrink.

Unfortunately, these stats don't yet account for the full browser's memory usage. The builds are debug anyway. I want to add total counts of JSObjects, and the size of the Cycle Collector graph.

This page is run with these prefs:

browser.dom.window.dump.enabled;true
dom.popup_maximum;700

and people.mozilla.com is allowed to open popups.
I should add, I also keep an altered copy of about:memory open during the run. It runs in a 10 second setTimeout, so I can track the stats over time.
Blocks: 631578
the WIP python analysis scripts. it processes the stdout/stderr of a debug browser build running membuster (or anything, I guess) and makes graphs.

http://hg.mozilla.org/users/rsayre_mozilla.com/membuster-analysis
Whiteboard: [MemShrink:P3]
Unfortunately the URLs in comment 0 no longer work because Sayre's page is gone.  The python scripts in comment 3 are still available.
This bug isn't useful any more.
Status: NEW → RESOLVED
Closed: 10 years ago
Resolution: --- → INCOMPLETE
You need to log in before you can comment on or make changes to this bug.