core-dump with aforce, TURBOMOLE v.6.0

[sassy]

Dear all,

I am having a problem getting aforce to run for a larg molecule (85 atoms, one Fe) using BP86/def2-TZVP.
The input file was generated with the latest version of the GUI and then copied over to the cluster. That worked well for
benzene example. The production run got that far (aforce.out):
 CONSTRUCTING second deriv. of 2e energy    -> Hessian
   treating Coulomb (and exchange) contribution
     integrals will be neglected if total contribution <
   2.5746652935118433E-010
      ...terminated. cpu:   87595.82       wall:   87595.82
   treating exchange-correlation contribution

and then generated a core dump due to a segfault:
aoforce[15165]: segfault at 000000004266d000 rip 0000000000be4905 rsp 00007fffffe7b4b0 error 6

As I am having 8 GB on that quadcore Opteron node, I decided to use:
$maxcor    1600
as aforce only runs on a single core and other people are using that node as well.

Would increasing the memory solve the problem or is there another issue here?

All the best

Jörg

[antti_karttunen]

Hello Jörg,

Segmentation faults are usually caused by too small stacksize limits. More info here: http://www.turbo-forum.com/index.php?topic=23.0

Regards,
Antti

[uwe]

Hi,

TmoleX can start Turbomole jobs on remote systems itself with the 'run network' button in the start-job panel. It generates a small script that is being executed on the target system. The script automatically includes the line

ulimit -s unlimited

This should help to get rid of a too small stack size limit which causes almost all segfaults.

Uwe

[sassy]

Hi all,

thanks for your feedback and sorry for my late reply.

I was actually using the start-job file from the GUI which contains the lines:

#!/bin/sh

rm -f Job_Ended
touch Job_Started
export TURBODIR=/usr/local/TURBOMOLE_6.0/
export PATH=$TURBODIR/scripts:$PATH
ulimit -s unlimited
export PARA_ARCH=MPI
export PARNODES=4
export PATH=$TURBODIR/bin/`sysname`:$PATH
aoforce >aoforce.out
#ridft >ridft.out
rm -f Job_Started
touch Job_Ended

So the ulimit -s unlimited was already set. I found that the easiest way to start the job.
I was starting the job manually (i.e. qrsh to a free node and then start it from the shell), so
the stacksize should have been unlimited.

Any other ideas?

Marry Christmas and all the best for 2010

Jörg

[uwe]

Hi,

Turbomole 6.0 - there was a problem with the AMD math library on some Opteron systems with the 6.0 version of Turbomole.

Please try either Turbomole 6.1 or contact the Turbomole support to get a fixed binary of 6.0 to check if your crash is related to this known problem.

Regards,

Uwe