I ran into a discussion on #sugar about a new matchbox rpm from tomeu with compositing enabled[1], and he asked me to email some testing results to sugar@ - in particular, to determine the memory impact, and whether that memory was in Xorg or the activities.
It looks like the memory is in Xorg[2], and is around 2 MiB per activity[3]: $ echo ; grep matchbox-window-manager: before after. ; for app in matchbox olpc-upd python Terminal Browse Write Journal Xorg ; do echo ----- ; grep $app before after. | grep MiB; done before:matchbox-window-manager: 1.2-3.20070628svn after.:matchbox-window-manager: 1.2-3.20070628svn.fc7 ----- before: 1.4 MiB + 2.8 MiB = 4.2 MiB matchbox-window-manager after.: 1.5 MiB + 2.8 MiB = 4.3 MiB matchbox-window-manager ----- before: 3.0 MiB + 2.3 MiB = 5.3 MiB olpc-update-que ----- before: 52.2 MiB + 10.5 MiB = 62.7 MiB python (6) after.: 40.6 MiB + 10.3 MiB = 50.9 MiB python (5) ----- before: 11.4 MiB + 10.6 MiB = 22.0 MiB Terminal <71a4d after.: 11.2 MiB + 10.6 MiB = 21.8 MiB Terminal <1d479 ----- before: 25.1 MiB + 11.0 MiB = 36.1 MiB Browse <955b6c4 after.: 25.2 MiB + 11.0 MiB = 36.2 MiB Browse <f54f896 ----- before: 16.6 MiB + 11.1 MiB = 27.7 MiB Write <e0a3864e after.: 17.3 MiB + 11.2 MiB = 28.5 MiB Write <76c06ca2 ----- before: 11.1 MiB + 10.6 MiB = 21.6 MiB Journal <38f01d after.: 10.8 MiB + 10.5 MiB = 21.2 MiB Journal <ff2f91 ----- before: 5.8 MiB + 1.7 MiB = 7.5 MiB Xorg after.: 16.9 MiB + 1.5 MiB = 18.4 MiB Xorg I hope I've done this right. I used ps_mem.py (per tomeu) and a wrapper script to gather these numbers[4]. Martin PS -- things feel a lot faster. 2 MiB/activity (of which I can generally get 4-5 usefully running at a time, and each of the ones tested takes over 20 MiB of memory, so I'm not sacrificing one potential activity I could be running) is fine by me - things feel a good amount faster. 1. at least I think that's the change: http://dev.laptop.org/~tomeu/matchbox-window-manager-1.2-3.20070628svn.fc7.i386.rpm 2. I could get Xorg to take up more memory by alt-tabbing as quickly as possible (would get up to 22 MiB, from 18). It's really easy to do this as alt-tabbing is now tons faster :). I could get it back down to 18.4 MiB by alt-tabbing through a few (at least 3, it seemed) activities, but slowly. Visiting any sugar shell screen(s) didn't seem to have this effect. 3. Going from 7 to 18 MiB = 11 MiB / 4 activities (Journal, Terminal, Write, and Browse) ~= 2.75 MiB / activity. If you count the shell as one activity (I think this makes sense if I understand what's going on here), that's down to ~= 2.2 MiB / activity. After launching etoys (which didn't seem to want to draw itself, but dunno if that's my build or the compositing) and stopping it, Xorg was down to 16.1 MiB. So I think there are some variables here I wasn't systematically testing, but the ~2 MiB / activity guideline seems useful for discussion. 4. The scripts and some raw outputs (I have much more, but I suppose people that care may find it quicker to re-do my tests than page through lots of my results) are: http://www.xades.com/proj/xo-psmem-matchbox/before http://www.xades.com/proj/xo-psmem-matchbox/after. http://www.xades.com/proj/xo-psmem-matchbox/olpc-psmem http://www.xades.com/proj/xo-psmem-matchbox/ps_mem.py
pgpFWr5bVB3O3.pgp
Description: PGP signature
_______________________________________________ Sugar mailing list [email protected] http://lists.laptop.org/listinfo/sugar
