Am Saturday 08 January 2011 schrieb oliverthered:
> Ok xorg was using a lot of ram
>                      mem          shared
> Xorg              729248k     9060
                     ^^^^^^^ "RES" or "VIRT" (anyway it seems too much)
the other values (iff "RES" are quite high as well)

> Here's the top few from xrestop
> ...
> res-base Wins  GCs Fnts Pxms Misc   Pxm mem  Other   Total   PID Identifier
> 1600000   103    1    0 1035 1966   197975K     48K 198024K  1257 kwin
runs @11767K here - there's probably a leak  - i assume you've composting 
active.
run "kcmshell4  kwincompositing", enter all "all plugins" and disable esp. the 
blur effect. if that doesn't help, disbale _all_, suspend/resume compositing 
(shiftt+alt+f12) and check whether the issue remains. if not, re-enable your 
effects one-by-one (Aspergers don't have a problem with this, everybody else 
runs 
grep -iE 'kwin4_effect_.*Enabled=true' `kde4-config --path config | cut -d":"  
-f1`/kwinrc | sed -e 's/kwin4_effect_//g; s/Enabled=true//g'
to dump a list of active effects ;-)

> cd /proc
> sudo grep -R --include='cmdline' --exclude-dir='[a-z]*'
> '.*graphicssystem.*' *
"unfortunately" the default value can be compiled into Qt - so it's not sure 
it's not the default (but from the xrestop output - probably not)
 
> Would it be worth just doing it when the system load if more idle i.e. is
> it better that it happens when nothing is happening or is it better that
> it's run when lots is happening.
the flush will block the framebuffer for a short moment, so rather do when 
nothings on the tasklist and also right after the desktop has completely 
loaded (the buffer's usually full&slow then)

> dirtywindows and the like (I would assume an attempt at doing as little as
> possible if nothing is happening in the window,
yes - the client causes damage events for "damaged" areas, those are handled.
converting x11 pixmaps to textures is expensive..

no idea how this interferes with vnc, never used =\

Thomas
 
>> Visit http://mail.kde.org/mailman/listinfo/kde-devel#unsub to unsubscribe <<

Reply via email to