As I type this the mem used is sitting at 245M after closing FRP. I restart it and FF3 drops back and is now at 150M. And it stays somewhat consistent varying only by 20M or so up and down. Might just be a quirk with my system but that’s what I’m seeing. Maybe someone can see the relationship (IF there is one) and figure out a consistent fix for everyone.

Ken

As for firefox, the paged memory should only exceed the private memory if privatememory runs low.

I increased my pagefile size (i just added the 2nd Gig a week ago) to 3069M with a max of 4500M. With the same number of tabs open to the same pages (though the content may be more or less active) my current mem usage is 154K and my VM size (which I wasn’t tracking before is 365M.

My processor usage by FF3 is lower as well though I can actually quantify the difference.

I also added FreeRamPro (a free download from download.net). Not sure if it has helped but it certainly hasn’t hurt.

Ken

Yes, I searched for “firefox memory hog” and yes, it’s true.

In the amount of time it took me to type that, I am now at 467,124 KB. And I type fast.

ie. save image -> download again.
view source (some websites) -> make duplicate request and resend unnecessary information, which could really screw up if you’re buying stuff online!!

a GDI memory leak can be easily destinquished in the GDI objects columb in task manager

http://www.microsoft.com/downloads/details.aspx?FamilyID=9b5edfc8-a4bb-4080-9063-6518166e2dab&displayLang=en

and after that, FF3 memory has been very stable. I’ve been running this about 3 or 4 days now, and FF3 memory use is OK. So maybe that will fix the problem. I would like to hear feedback from others.

i created a second test profile just for this lol.

Fresh start with bfcache disabled, is 49MB’s at this point.

Ok, using 10 sites opening them twice each.

forums.ngemu.com, redcl0ud.1.forumer.com, onemanga.com, forum.pj64-emu.com, http://www.jabosoft.com, forum.nhancer.com/, http://www.msghelp.net/, http://www.ngohq.com, community.mybboard.net and mail.live.com

thats about 77MB’s used.

right click a tab and click close others

this dials down to 72MB’s.

Open the same tabs again, takes us up to 89MB’s

Closing them again takes us down to 84MB’s

Opening those same Tabs again takes us to 96MB, and closing them takes us down to 90MB.

Ok, at this point, im at 90MB, im going to open the same tabs 4 times each, so thats 40 open tabs.

Theres 125MB’s. closing them takes us back to 103.

Now this was a controlled example, obviously we aren’t going to be loading the same sites 4 times each, and data and images differening in each tabs is going to hit on memory usage much more.

Start with one tab open (call this low number L1)
Open ten tabs (and call this higher number H1)
close the ten tabs (call this L2)
open ten tabs again (call this H2)
etc.

Measuring memory usage from L1 to L2 to L3, etc., is going to tell you how good the implementation of malloc and free is at returning memory to the system, how fragmented the allocations end up, etc. It’s not clear if this graph is very meaningful to what users experience.

Measuring memory usage from H1 to H2 to H3, etc., is much more likely to yield useful information, and I think is more likely to reflect whether those users who aren’t looking at task manager are actually perceiving memory use problems.

I’ve loaded a session using session manager extension and saved them all as bookmarks, so i could use it in the fresh profile

Clean profile fresh start w BFcache disabled, is 51MB’s.
Load all the tabs at once(44), takes me up to 236MB’s

Close all tabs except for 1, memory usage dials back to 170MB, again.

Having used firefox 2.0 heavily previously, 50 or so tabs would use 240ish MB’s, and with BFCache set to 1 would dial back to 125MB’s, Opening a few tabs with a google search in them then closing them would at this point dial memory back to 90MB’s.

My own theory on it is that the memory used by images in a tab is not being released as it is more noticeable when sites with many images, or large hi res images are closed. I’m going to raise the issue at mozillazine and get some good reprosteps down, i’ll also build a places.sql file which has sites which agravate the problem.