618034 - Investigate memory increase that happened in Sept 2009 as identified in bug 598466

Status: RESOLVED INVALID

(Core :: General, defect)

Description

[Sheila Mooney]

This bug is for memory increase D as described in detail in. Per dmandelin's request we are adding a bug to investigate further why this happened, if the increase was intentional (expected tradeoff from other work) and if there are any ideas to mitigate it.

Comment 1

[Joe Drew (not getting mail)]

This should have gone away when bug 563088 landed. Is that not the case?

Comment 2

[The 8472]

see bug  bug 598466 comments 57, 61 and 53. The tests were conducted with image.mem.discardable = false

Therefore neither the increase nor the decrease related to image discarding should show up in these tests.

so there are two possibilities:
a) something messed up our testing procedure (prefs changing?)
b) the regression is unrelated to image discarding

Explicit tests showed that image discarding is working and yields similar results as it did in 3.6 (see bug 598466 comment 35)

Comment 3

[Joe Drew (not getting mail)]

In that case I think we need to rediagnose this bug, unfortunately.

Here's the regression range, as found by Ed in bug 598466 comment 63: http://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=bf0fdec8f43b&tochange=912c6ae3b70c

Comment 4

[Boris Zbarsky [:bzbarsky]]

> The tests were conducted with image.mem.discardable = false

That pref had no effect during the time of the observed regression.  It was added much later.  (I sort of said that in bug 598466 comment 65, but wasn't sure; I just verified that this is the case.)

At the time, you could disable image discarding using the MOZ_DISABLE_IMAGE_DISCARD environment variable.

Comment 5

[Ed Morley [:emorley]]

Thanks for breaking this out and ensuring it got followed up, I was going to go through bug 598466 and follow up the loose ends at some point, just hadn't had a chance.

Given bug 598466 comment 65, my feeling is that this increase might be due to the discardable about:config pref not existing back then. When I tested I only used the about:config setting and didn't set the environment variable, so I will retest shortly with the environment var set ("SET MOZ_DISABLE_IMAGE_DISCARD=1" in the command prompt prior to running yeah?) and see if that affects (effects?) the results.

Comment 6

[Ed Morley [:emorley]]

Ah ok, missed comment 4 before writing that, somewhat made my previous post redundant. Well guess it's more than likely the ENV variable is the cause then, but will confirm for sure tomorrow. Thanks!

Comment 7

[Joe Drew (not getting mail)]

Ed, have you had a chance to look at this?

Comment 8

[Ed Morley [:emorley]]

Sorry for the delay.

Using the _90_ tab testcase and methodology (ie: discardable about.config pref set to false, but environment variable not set) from bug 598466 comment 97:
2009-09-13: 330/348/495
2009-09-14: 508/530/667

Now with environment variable MOZ_DISABLE_IMAGE_DISCARD=1:
2009-09-13: 515/536/684
2009-09-14: 500/522/649

Therefore this increase was purely due to the image.mem.discardable = false preference not being implemented in builds before 2009-09-14 - and the need to therefore use the environment variable instead.

Marking as resolved invalid, sorry for the false alarm.
(Least one less increase to deal with now!)