Re: Good tests for real-world browser memory usage?

Here's an even simpler case:


http://www.tripadvisor.com/HotelDateSearch?d=607162&returnTo=__2F__Hotels__2D__g187514__2D__Madrid__2D__Hotels__2E__html&fromPop=false&from=HotelsButton

(make sure you allow pop-ups)

1) Load above URL
2) Click on 'check rates'
3) allow all windows to open
4) close them all except 1 and go to about:blank

On Vista (Working Set):

IE7:35MB->150MB->118MB
Minefield:35MB->118MB->94MB  (88MB after "clear caches" with Stuarts 
extension)
Safari3:25MB->140MB->110MB

Were does all the mem go between 3 and 4?

PS if you repeat this several times the max mem # continues to rise.

Cheers,

Schrep



Mike Schroepfer wrote:
> Hey Folks,
> 
> Thanks to the really hard work of DBaron, Sayre, Rob, Alice, Graydon and 
> countless others I'm forgetting we've got much better visibility into 
> Firefox memory usage than ever before.   In particular we tools to find 
> memory leaks and through Talos we can measure memory usage during a Tp 
> run (loads ~2000 pages) both between builds:
> 
> http://graphs.mozilla.org/#spst=range&spstart=1180657203&spend=1191603512&bpst=cursor&bpstart=1180657203&bpend=1191603512&m1tid=9&m1bl=0&m1avg=0 
> 
> 
> and during a test run:
> 
> http://graphs.mozilla.org/dgraph.html#spst=range&spstart=0&spend=2001&bpst=cursor&bpstart=0&bpend=2001&name=tp_Working%20Set&m1tid=29127&m1bl=0&m1avg=0 
> 
> 
> During these runs trunk maxes out at around 90MB of working set size on 
> trunk and 70MB on 1.8 branch.   However, many of the complaints we
> get about Firefox memory usage are when it gets much higher than this (> 
> 150MB).   Also a really quick and dirty informal test logging into 
> gmail, google reader, and loading cnn.com on a 2GIG Vista system I get:
> 
> IE7: 86.5MB
> Safari3: 78.7MB
> Trunk: 63.3MB
> Firefox2.0.0.7: 53.9MB
> 
> The leads me to a bunch of questions:
> 
> 1) The number reported on the Talos tests is the number at the end of 
> the run.  Should this number be the peak instead (or in addition)?
> 
> 2) Looking at the second graph above you see memory usage settle into a 
> a sine-wave like pattern with 4 or so local maximum's spaced about 5 
> minutes apart.  The peak of these goes up about 700KB each time (not 
> much given the number of pages loaded).   Should we expect these 
> maximums over a long enough run to stabilize (e.g. never increase beyond 
> a certain value)?
> 
> 3) Is there a user-initiated way to purge all caches (esp BFCache)?
> 
> 4) These tests are clearly do not reproduce the conditions causing 
> people to have much larger memory use.  Obviously missing is user 
> interaction, XMLHttpRequest, and other interactive elements.  This is 
> also a clean profile with no extensions.  So any ideas on how we can 
> build a more real-world test suite such as:
>    * Many more page loads
>    * Playback of user events
>    * Running with top 10 extensions
>    * Running Mochi or other testsuites
> 
> If we tried to construct a more "real-world" user-event driving test 
> what do people think is both representative and useful?  E.g. logging 
> into gmail, search for a message, log into flickr, do stuff, etc... Even 
> spot testing by hand key builds of FF might give us a useful measuring 
> stick.
> 
> Best,
> 
> Schrep
> 
> 
> 
> 
> 
> 
>