Remote Access + Hardware acceleration

[JaXXoN]

Originally Posted by junksiu

Are the server and development machines on the same LAN or remote like work from home?

I just tried out video playback with gigabit ethernet and i guess it wouldn't work
with the default vnc server when using DSL or a cable modem. But VirtualGL might
be good enough (however, i didn't tried it out this way, yet). AFAIR, the VGL VNC server
also runs in background (means: the user in front of the machine won't recognize it),
but i don't know if it possible to run several instances of VGL ...

regards

Bernhard

[Stephen Warren]

I briefly took a look at the NX website last night and some other information about it. NX does appear to completely replace the X server (it runs on the cilent machine) and hence you can't use VDPAU.

It's possible we have some professional/Quadro solutions that might help you out. I'm attempting to track down the best contact for you to speak to. I'll PM you the information when I find it.

[bdmw1]

Thanks Stephen. I will try out your recommendation once I hear back from you.

Also, is it true that there ain't any consumer grade/open source remote access program I could use which provides good speed (better than VNC, closer to NX) and still allow the use of VDPAU?

[junksiu]

Also check out diskless boot client while the client machine boot over the LAN and all data is on the server.

http://www.faqs.org/docs/Linux-HOWTO...ess-HOWTO.html

Some custom startup script will be needed to use a single image.

I have setup client machine with nvidia card running nvidia driver with full graphical acceleration like compiz. And since all program is running locally in the client machine, you use the full power of that box. The server box is only acting as a nfs file server. You may still want to put in a small hd in the client machinefor swap file if developer are doing huge compilation. With 2G ram client machine I can compile the linux kernel without swap.

[etherfish]

Assuming question #2, the answer is yes. The solution requires no extra software. No VNC, NX or anything else. In my computer science lab, we use this method all the time to put video, webpages, etc. up on a projector in our lab from our workstations.

All you need to do is on your server, set the DISPLAY environment variable to "display-client-hostname:0.0"

For example, in our lab, we call the computer attached to the projector "mruniverse" so, to redirect X-windows output from my computer to the projector computer, in a shell, I run:

export DISPLAY=mruniverse:0.0

If you don't have DNS, you could easily use the IP address. It's most likely that your linux distribution has network accces to the X server disabled. Enabling it properly is a bit complicated, but here's a quick way to manually start the X server to work well in this mode. First, you have to shutdown any graphical login program, like gdm or xdm.
Try these two:

/etc/init.d/gdm stop
/etc/init.d/kdm stop

You should now be at the white-on-black, all-text Linux Console screen. Log in as root and type:

X -noreset -ac &

X windows will start and you will have a cursor, but you will not be able to do anything. Literally. Don't worry. If you need to shut down the server press Ctrl, Alt, and Backspace simultaneously.

On another machine, type:

export DISPLAY=IPofYourDisplayComputer:0.0
xterm &

You should now see on your display an xterm window. There is no window manager running, so don't expect to be able to move the windows around or resize them or anything.

You should try running a movie player like mplayer to see if that will render your h264 content well enough. The XV extension will work properly over the network, btw. I can give you more details if you'd like.

[etherfish]

Here's a setup we use for watching the two cameras using some old LCD's that we'd run out of room to store. This uses the remote X11 xvideo as I described above.

All 12 displays are run by one machine sitting behind the LCDs. It is a Dual P4 Xeon, 2.8GHz, w/ 1GB of RAM. It has an AGP GeForce 5200 and five PCI GeForce 5200s. It's an Xinerama desktop composed of six twinview'd screens. In this picture, both of the camera feed's being displayed are an mplayer process on another machine playing a v4l2 device and the video out is XVideo via remote X11 over gigabit ethernet. The left camera is running at 15fps (usb webcam limitation) and the right one is 30fps (real tv camera attached to video framegrabber.) Both streams are 640x480 scaled up (by the graphics cards) to 2048x1536. We don't have any perceptible frame drop. We did find that using Jumbo frames (9000 byte MTU) on the gigabit ethernet reduced the load considerably. We specify on the mplayer command line the geometry of the window it makes.

e.g, for the first camera:

export DISPLAY=hydravision:0.0
mplayer -screenw 4096 -screenh 2304 -xy 2048 -geometry +0+0 <more mplayer stuff>

and for the second camera:

export DISPLAY=hydravision:0.0
mplayer -screenw 4096 -screenh 2304 -xy 2048 -geometry +2048+0 <more mplayer stuff>

The -geometry and -xy lines, specifically, specify the size of the mplayer window and it's location on the screen.

[Stephen Warren]

VDPAU won't work across remote X11 protocol; it's a so-called "direct rendering client", and hence can only display video on the machine upon which the program is running.