-
Notifications
You must be signed in to change notification settings - Fork 883
DirectFB: Improve performance of event handling #44
Comments
Take a look on wait_obj. It might be what you're looking for. |
Just looking around for running loops in the tree: $ grep -RHn 'while (1)' * channels/rdpdr/smartcard/scard_operations.c:461: while (1) channels/rdpdr/smartcard/scard_operations.c:660: while (1) channels/rdpdr/rdpdr_main.c:827: while (1) channels/rdpdr/rdpdr_main.c:896: while (1) channels/rdpdr/rdpdr_scard.c:78: while (1) channels/rdpdbg/rdpdbg_main.c:112: while (1) channels/rdpdbg/rdpdbg_main.c:157: while (1) channels/rdpsnd/rdpsnd_main.c:203: while (1) channels/rdpsnd/rdpsnd_main.c:581: while (1) channels/rdpsnd/rdpsnd_main.c:632: while (1) channels/drdynvc/drdynvc_main.c:462: while (1) channels/drdynvc/drdynvc_main.c:508: while (1) channels/drdynvc/audin/alsa/audin_alsa.c:199: while (1) channels/cliprdr/cliprdr_x11.c:892: while (1) channels/cliprdr/cliprdr_main.c:189: while (1) channels/cliprdr/cliprdr_main.c:234: while (1) dfb/dfbfreerdp.c:566: while (1) dfb/dfbfreerdp.c:743: while (1) libfreerdp-core/mppc.c:334: while (1); libfreerdp-core/mppc.c:368: while (1); libfreerdp-gdi/gdi_32bpp.c:1086: while (1) libfreerdp-gdi/gdi_8bpp.c:844: while (1) libfreerdp-gdi/gdi_16bpp.c:1013: while (1) win/wfreerdp/wfreerdp.c:541: while (1) win/wfreerdp/wfreerdp.c:712: while (1) X11/xfreerdp.c:694: while (1) X11/xfreerdp.c:872: while (1) If these are all running continuously in the different threads they will definitely be a battery killer on any portable device. . |
We surelly welcome patches to address those :-D |
I'm sure. :-) I think this one is going to take a team effort to look at this. The scope is pretty big and you would have to do all of it in order to accomplish the goal. . |
Yep, just what I was afraid of: PID USER PR NI VIRT RES SHR S %CPU %MEM TIME+ COMMAND 9979 root 20 0 446m 20m 2920 S 21.6 0.3 0:06.60 lt-dfbfreerdp This is just at idle with no plugins, no user activity, no server activity. That level of idle CPU usage will kill a battery on a portable device in short order. |
Yes ... I fully agree. Surely this is a team-work and we ought to commit to avoid this to happen again in the future however it is natural to have those issues as the project start to mature itself. Currently I am quite busy on other stuff so I won't be at help for now. If you can start looking at it and propose patches it would be awesome. I am sure that others can help with this issue and also help to review proposed patches to address these issues. |
Right now it looks like these loops are all "polling" loops. I think what we need are "wait event" loops that wait for an event interrupt to occur before they run the next cycle. These are the copyrights from the files uniquely sorted, so probably they should make up the team I would think. Copyright 2009-2011 Jay Sorg Copyright 2010-2011 Vic Lee Copyright 2010 Marc-Andre Moreau Copyright 2011 Eduardo Fiss Beloni Copyright 2011 O.S. Systems Software Ltda. Copyright (c) 2009-2011 Jay Sorg Copyright (c) 2010-2011 Vic Lee Copyright (C) Alexi Volkov 2006 Copyright (C) Jay Sorg 2009-2011 Copyright (C) Jay Sorg 2010-2011 Copyright (C) Matthew Chapman 1999-2008 |
Some of those do that. You can even see how it is done to try to accommodate it your the one you wish. |
Ok, I just reconfigured to compile X11 version. It is showing less cpu usage than dfbfreerdp version. So what is needed to make dfbfreerdp to behave like xfreerdp? Maybe it is that improvements for event handling did not get migrated into the dfbfreerdp UI. I'll try to look at this some more maybe tomorrow. Maybe Jay or Vic could chime in here. |
I noticed that the event handling is setup as a running loop in freerdp.
That works fine for workstations and PC's that have unlimited power sources.
But for laptops and embedded devices running loops can be battery killers.
Maybe we could explore some type of wait_event mechanism for event handling.
I stuck in a DEBUG statement to see what was going on and I could see that the loop was going round n' round all the time even though my hands were off the keyboard and mouse and there was nothing happening on the server.
The text was updated successfully, but these errors were encountered: