[Sugar-devel] [RELEASE] Surf 106

[Bobby Powers]

On Wed, Mar 4, 2009 at 1:01 PM, Jonas Smedegaard <dr at jones.dk> wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> On Wed, Mar 04, 2009 at 11:27:00AM -0500, Bobby Powers wrote:
>>On Wed, Mar 4, 2009 at 6:42 AM, Jonas Smedegaard <dr at jones.dk> wrote:
>>> On Wed, Mar 04, 2009 at 09:21:50AM +0000, Bobby Powers wrote:
>>>>An informal test showed that Browse in sugar-emulator used 100MB in
>>>>opening and navigating to gmail, while surf used 85MB.  That still
>>>>seems like a lot, but its a 15% savings right off the bat.
>>>
>>> Please document how to measure the memory use (even if non-academic),
>>> to make it possible to compare on other environments using exact same
>>> measuring method (as I suspect it may vary a lot, depending on
>>> compile options of e.g. xulrunner).
>>
>>What I did:
>>step 1: run a new instance of sugar-emulator
>>step 2: click on the browse or surf icon from the home view
>>step 3: navigate to https://mail.google.com
>>step 4: log in using your gmail credentials
>>step 5: open gnome-system-monitor and check the memory usage
>
> Thanks.
>
> You did mention that your test was informal. Anyway, here are some notes
> if someone ones to test further:
>
> I avoid registered Google services, and I guess I am not the only one:
> It would probably be good to use some public web pages.

I will do more testing over the next few days.  My choice of gmail was
based on the fact that its one of the most complex web
sites/applications I could think of quickly.

> Memory usage seem to also be about cleaning up memory[1], so probably
> would make sense to measure a larger number of web pages (fully loading
> all content on each page and then move on to the next page).
>
> I don't use GNOME, and don't know how it calculates memory consumption.
> Probably would be better to use terminal-based measurement like "free"
> or some other tool providing more optimal info.
>
> While googling for info about this, I also stumbled across a hint that
> Gecko-based browsers can be configured to use more or less memory cache
> - - and has a live status by entering about:cache in the address entry.
> Would be interesting to know if WebKit-based browsers have something
> similar.

I think there are configuration options somewhere, as I seem to
remember that Chrome on Andriod (based on webkit) can be configured to
clear its memory caches when the kernel signals that it is low on
memory.

>
>  - Jonas
>
>
> [1] Pages like http://dotnetperls.com/Content/Browser-Memory.aspx
> demonstrate how WebKit-based browsers in the past have struggled with
> not cleaning up properly after itself. That test was done on Windows,
> but http://blog.pavlov.net/2008/03/11/firefox-3-memory-usage/ mentions
> how recent Gecko at least (don't know about WebKit) use same memory
> allocator on Windows and Linux.

wow.  it should be quite interesting to look at the memory usage over
time to see if webkit has either fixed their memory leaks or caching
strategy.