Difference between revisions of "Problems and Solutions on SPL Machine Blog"

From NAMIC Wiki
Jump to: navigation, search
 
(15 intermediate revisions by the same user not shown)
Line 3: Line 3:
 
----
 
----
  
−
With the install of Fedora 7 we have been having some problems, namely with the OpenGL driver recognition by SCIRun.
+
Jan 08:  There are currently no SPL specific problems with SCIRun on SPL machines.  There is a more general bug in SCIRun related to questionably threadsafe code, specifically DLOpen calls which are primarily in the dynamic compilation portion of SCIRun.  These show up often on the fat nodes. Jeroen is working to eliminate dynamic compilation, and eliminate these bugs which manifest more frequently  and randomly with large networks and multicore machines which "stress" the thread safety of code.  If you are using SCIRun and running into these bugs please let him know.
−
 
 
−
For example when we utilize the tools on the previous page SCIRun gets pointed at the correct drivers as shown here:
 
−
 
 
−
spl_tm64_1:/workspace/mjolley/Modeling/trunk/SCIRun/bin% ldd scirun | grep GL
 
−
        libGL.so.1 => /usr/lib64/nvidia/libGL.so.1 (0x0000003f8e200000)
 
−
        libGLU.so.1 => /usr/lib64/libGLU.so.1 (0x000000360c200000)
 
−
        libGLcore.so.1 => /usr/lib64/nvidia/libGLcore.so.1 (0x0000003f80e00000)
 
−
 
 
−
 
 
−
When these drivers are recognized SCIRun runs as expected.  However, each time SCIRun is run it reverts back to the wrong drivers as evidenced here.
 
−
 
 
−
spl_tm64_1:/workspace/mjolley/Modeling/trunk/SCIRun/bin% ldd scirun | grep GL
 
−
        libGL.so.1 => /usr/lib64/libGL.so.1 (0x0000003f84800000)
 
−
        libGLU.so.1 => /usr/lib64/libGLU.so.1 (0x000000360c200000)
 
−
 
 
−
After this, any OpenGL dependent modules crash upon opening.  If you repeat the steps "unsetenv" and run Dav's script again you get back to:
 
−
 
 
−
 
 
−
spl_tm64_1:/workspace/mjolley/Modeling/trunk/SCIRun/bin% ldd scirun | grep GL
 
−
        libGL.so.1 => /usr/lib64/nvidia/libGL.so.1 (0x0000003f8e200000)
 
−
        libGLU.so.1 => /usr/lib64/libGLU.so.1 (0x000000360c200000)
 
−
        libGLcore.so.1 => /usr/lib64/nvidia/libGLcore.so.1 (0x0000003f80e00000)
 
−
 
 
−
To summarize:
 
−
 
 
−
 
 
−
unset LD_LIBRARY_PATH
 
−
ldd scirun | grep GL
 
−
<correct GL is reported>
 
−
run script
 
−
ldd scirun | grep GL
 
−
<wrong GL is reported>
 
−
 
 
−
 
 
−
So I put:
 
−
 
 
−
unset LD_LIBRARY_PATH into the first line of the script(run in bash) and into my .bashrc and I still have the same behavior where it switches back from the good OpenGL setup to the mesa drivers after initially being pointed to the correct ones.
 
−
 
 
−
This still didn't seem to do it so we put in the following line in the script:
 
−
 
 
−
create_scirun_script() {
 
−
    echo "scirun -E ${NETWORK} --logfile ALL.log" >/tmp/script-fe.sh
 
−
    chmod 0770 /tmp/script-fe.sh
 
−
}
 
−
 
 
−
TO:
 
−
 
 
−
create_scirun_script() {
 
−
    echo "unset LD_LIBRARY_PATH" > /tmp/script-fe.sh
 
−
    echo "scirun -E ${NETWORK} --logfile ALL.log" >> /tmp/script-fe.sh
 
−
    chmod 0770 /tmp/script-fe.sh
 
−
}
 
−
 
 
−
And the above behavior with it switching back to the MESA drivers appears to have stopped, but the script still always hangs at JoinField in when the third net is run:
 
−
 
 
−
 
 
−
 
 
−
That said it still will not run past the third net.  I am going back an confirming this set of .bdl files runs on my ubuntu machine.
 
−
 
 
−
 
 
−
 
 
−
.--[[User:Mjolley|Mjolley]] 13:38, 20 August 2007 (EDT)
 

Latest revision as of 19:14, 28 January 2008

Home < Problems and Solutions on SPL Machine Blog

Current Problems on Debugging for SCIRun on SPL Machines

Jan 08: There are currently no SPL specific problems with SCIRun on SPL machines. There is a more general bug in SCIRun related to questionably threadsafe code, specifically DLOpen calls which are primarily in the dynamic compilation portion of SCIRun. These show up often on the fat nodes. Jeroen is working to eliminate dynamic compilation, and eliminate these bugs which manifest more frequently and randomly with large networks and multicore machines which "stress" the thread safety of code. If you are using SCIRun and running into these bugs please let him know.

Retrieved from "https://www.na-mic.org/w/index.php?title=Problems_and_Solutions_on_SPL_Machine_Blog&oldid=21569"