I aggree with -C Wayne Wright 

I just don't run ANY X stuff on an RT-Linux CPU.

When I started my RTLINUX CNC development I made the decision early on to keep all RT 
control on a 
single CPU that would run RT-LINUX (NO X server) and control the CNC hardware.  The 
GUI is running on a neighboring (networked) NON-RT linux box which runs it's own 
xserver.  ALL Time critical stuff runs on the RT linux box.  The GUI and the Xserver 
is running on the NON RT BOX.  This organization serves several purposes:

1) IF I Crash RT-Linux due to bad memory acceses (in a shared memory RT application) I 
do not have to re-boot
my development workstation.  Since I am just learning, this has been more of a help 
than I first thought :) The 
smaller RTlinux box reboots quicker since it is not running X and other stuff.  

2) Timing on the RT machine is not clouded by jitter from the X server, big disk 
accesess,  and other processes.  There is a bare minimum of software running on the 
RTcontrol CPU.  If I crash or have a timing problem, I can find the cause much more 
quickly.

3) You DON'T need as much CPU horsepower on the control machine since there is less 
running on it. I am using a 486DX100 and find that I can easily run repeative tasks at 
200us intervals.  These machines are also CHEAP. I gor mine for FREE from a friends 
basement. 

4) I can rebuild the RT-Linux kernel ANYTIME without interfering with my editors, 
files etc.  I don't have to worry
about breaking my other applications when I reboot the RT-Linux kernel.

5) I NFS mount the development directory from the GUI computer onto the RT-Linux box.  
This places all critical 
files for development in one place.  I compile the rt-linux code on the rtlinux box 
and the GUI code on the GUI box.

6) I can move this all onto a SINGLE RT-Linux box AFTER devlopment is complete.  Then 
use the other CPU as 
a boat Anchor!!


Only the guys who REALLY know what they are doing can keep an RT-Linux box from 
crashing while they
are developing their real time code !!! It is way to easy to make a serious mistake in 
the real time code!!

dave
Building CNC control software in TCL/TK on linux and RT-liunx

see the MONSTER CNC mill at
http://users.nni.com/daveland/metal_madness.htm

C.W.Wright Wrote ........

Run "top" when you're executing the test.  I found that almost all of the
overhead in my realtime linux/X systems is the X server.  When the app
code and the X server are on the same machine, and I update "waveform
displays", the X server is the top cpu hog, and the driving code (client)
is very low on the list.  I have display apps in both "C" and "Tcl/Tk/Blt"
and even the Tcl/TK code uses only a few percent of the CPU while the
Xserver can easily use 60-80%.

My partial solution was to run the Xserver on a networked laptop running
the Xserver as an X terminal with:

    X -query realtime-machine.net

With:

   X :1 -query realtime-machine.net

You can have one Xserver for the laptop, and the second one taking care of
the realtime-machine.net.

RT/linux is great!!!!!!! :-)

-C Wayne Wright



--- [rtl] ---
To unsubscribe:
echo "unsubscribe rtl" | mail [EMAIL PROTECTED] OR
echo "unsubscribe rtl <Your_email>" | mail [EMAIL PROTECTED]
----
For more information on Real-Time Linux see:
http://www.rtlinux.org/~rtlinux/

--- [rtl] ---
To unsubscribe:
echo "unsubscribe rtl" | mail [EMAIL PROTECTED] OR
echo "unsubscribe rtl <Your_email>" | mail [EMAIL PROTECTED]
----
For more information on Real-Time Linux see:
http://www.rtlinux.org/~rtlinux/

Reply via email to