|
|
#13 | |
|
Registered User
Join Date: Feb 2009
Posts: 138
|
Uhm, why haven't you tried 256.53? Note the changelog.
|
|
|
|
|
|
|
#14 | |
|
Moderator
Join Date: Aug 2005
Posts: 1,327
|
FYI, this is the test script I used:
for i in `seq 10`; do for j in `seq 100`; do (glxgears&); sleep 1; kill `pidof glxgears`; done; sync; sleep 10; for x in `seq 3`; do sudo sh -c "echo $x > /proc/sys/vm/drop_caches"; done; sleep 10; grep 'kB' /proc/meminfo |cut -d: -f2|sed -e 's/kB//' -e 's/ //g' |xargs echo >> ~/shared/ion-memusage-flushing-cache.txt; done This removes disk cache effects from measurements. I also tried "xclock" instead of "glxgears" and was able to repro the problem. I could only repro with full Gnome running; when I start a bare X server and a single xterm (to prevent resetting the X server every time the client disconnects) I don't see the issue. I was seeing Free memory fall, while various columns relating to "anonymous mappings" were increasing. This all implies that some Gnome application is causing the problem. However, I wasn't able to track down how or why yet. Does that match your results? |
|
|
|
|
![]() |
| Thread Tools | |
|
|