[Testing] [Fwd: [support-gang] simple hacks to improve the performance of the Sugar UI]

Mel Chua mel at laptop.org
Wed Nov 5 15:48:04 EST 2008


Just for fun (read: totally not high-priority - Activity testing is 
still our most important work), but throwing this out here in case 
anybody feels like testing on some of Erik's performance work (potential 
side bonus: your XO should "feel" way, way faster.) Feedback should go 
to Erik Garrison <erik at laptop.org>, or erikg on IRC.

NB: At some point during our weekly meetings, we should probably figure 
out a good way to handle these "developers have made something and need 
people to try it out!" requests - tomorrow's agenda looks pretty 
stuffed, but I've placed it on the agenda queue, 
http://wiki.laptop.org/go/Testing_meetings/Agenda_queue.

-Mel

-------- Original Message --------

XO Users,

In short, I have bundled a set of 'hacks' which generally (and in my
observation, dramatically) improve the user-perceived responsiveness of
the Sugar UI.  The hack bundle is available at
http://dev.laptop.org/~erik/faster-hacks.zip

I write requesting independent evaluation of the costs and benefits of
these modifications.

To test you *must* be running our current official release: 8.2-767.



-== Contents ==-

There are several 'hacks' included in this bundle:

 1) optionally disable frame appearance on activity switching, if the
file /home/olpc/no-frame-on-tabbing exists

 2) optionally disable autosaving on activity switching, if the file
/home/olpc/no-auto-save exists

 3) optionally start the X composite manager (xcompmgr) on sugar startup
to prevent the redraw of activity windows when switching from one
activity to the other (if /home/olpc/use-xcompmgr exists).

 4) compcache, http://code.google.com/p/compcache/, which makes a
compressed block device usable for swap out of a chunk of RAM.  By
compressing unused pages this system may offset memory usage increases
incurred by using composite.


-== Usage ==-

To install and test on an XO, open the Terminal activity and run:

    wget http://dev.laptop.org/~erik/faster-hacks.zip
    unzip faster-hacks.zip
    cd faster-hacks
    sudo ./install-faster-hacks

Restart Sugar, then, if you'd like to test with compcache, run:

    cd faster-hacks
    sudo ./start-compcache

Note also that you can adjust the amount of RAM (in kilobytes) devoted
to compcache by passing an argument to ./start-compcache, e.g.:

    sudo ./start-compcache 50000

By default it's set at 100000Kb (~100Mb).


You can enable hacks 1-3 while running Sugar by executing:

    ./enable-faster-hacks

and disable with:

    ./disable-faster-hacks

Effects will be immediately apparent.


The changes in Sugar source files and installed executable (xcompmgr)
can be reverted and removed with:

    sudo ./uninstall-faster-hacks

This step is a failsafe and likely unnecessary; ./disable-faster-hacks
should be sufficient to stop the effects and the size of the installed
programs is negligible (<30kb in xcompmgr).



-== Discussion ==-

Now continue to play with your XO as normal!  Please report experiences
and impressions.  The 4 hacks are independent in operation and may be
mixed and matched to achieve a variety of possible configurations.

Keep in mind that these hacks present a tradeoff between existing
features and UI responsiveness!

The disabling of the frame removes a feature, but this feature (frame
auto-appearance) may be unnecessary if we can rapidly switch between
activities, as xcompmgr enables us to do.

The disabling of autosaving removes a feature, but it drops the load on
the processor which we incur when navigating between activities, thus
improving switching performance.  

The use of xcompmgr represents a tradeoff between free RAM and system
responsiveness.  RAM is utilized to store offscreen framebuffers for
each window in the X server.  This obviates the need to have programs
redraw themselves when they are switched into, dropping CPU usage during
switching.

But part of the RAM hit we incur by running xcompmgr may be offset by
using compcache to provide a compressed swap partition-in-RAM.

Comments more than welcome!

Erik





More information about the Testing mailing list