Re: [GnomeMeeting-list] Unlimiting Bandwidth?



On Tue, 2003-05-20 at 10:06, Damien Sandras wrote:
> Le mar 20/05/2003 à 16:59, Jeffrey C. Ollie a écrit :
> 
> > I'm using a nVidia GeForce2 MX 400 with the binary XFree86 drivers from
> > nVidia.  xvinfo confirms that YUV overlays are supported.  My camera is
> > a Logitech QuickCam Pro 4000.  Is there some way of determining for sure
> > that YUV overlays are being used?  During a call XFree86 is using up a
> 
> Yes, there exist an environment variable to set to put SDL in Debug
> mode. Try to switch to full screen and watch the CPU usage, it should go
> down.

Ok, I found the environment variable, but it prints nothing.  Messing
with some of the other variables does nothing either.

<ftp://ptah.lnf.kth.se/pub/misc/sdl-env-vars>

> > lot of CPU too.  Here's the output from "-d 6" that relates to the video
> > grabber:
> 
> Just for curiosity, what kind of CPU do you have? 

1.8Ghz Pentium 4.

One other interesting point... I set up a videoconference directly
between GM and one of the PolyCom end units.  CPU usage was dramatically
lower on the point-to-point than when I was connected to the Cisco MCU. 
I might have to try and set up OpenMCU tomorrow.  Maybe the Cisco MCU is
doing something that makes GM have to work harder.

I'll have to put GM in debug mode tomorrow and take a look at the H.323
setup messages that are being exchanged.

> > Yes, this definitely stresses the CPU.  Is this implemented in GM or in
> > OpenH323?  Too bad I don't know x86 assembler or I'd try rewriting
> 
> In GTK directly.

Hmmm... If it's in GTK I wonder if it has been optimized very much. 
It'd be interesting to compare the version in GTK vs. similar code in
one of the open source video processing libraries, like transcode or
something like it.

> > subroutines like these to see if there would be performance benefit.
> > 
> > I also noticed that reducing the size of the transmitted video from
> > large to small had a significant effect on CPU usage.
> > 
> 
> Yes, actually the images have to be converted from one color format to
> the other for display/reception/transmission. But the CPU usage
> shouldn't go to 100% except perhaps on Pentium1 machines.

Yeah, wierd...

Jeff





[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]