The problem has been solved by changing a setting on the head node so
that the error is no longer displayed.
Thanks,
Jennifer Thorne
On Wed, 2010-06-23 at 08:34 -0400, Jennifer Thorne wrote:
> Hello,
>
> I am running an optimization software from Sandia Labs called DAKOTA,
> this software calls simulator_script which runs a simulation called
> GAMES in parallel on a cluster. I first call lamboot with a hostfile
> and then call DAKOTA which calls the simulator_script which sets up
> the parallel computing and runs GAMES with run_Gamesv1p2.c which calls
> the shell script that runs the Matlab GAMES simulation.
>
> My problem is that LAM receives the following message on stderr xhost:
> unable to open display "" when it calls /usr/bin/ssh -x -a head -n
> 'echo $SHELL' on the nodes other than head in the cluster. The command
> correctly prints $SHELL but first displays the xhost error.
>
> I've tried the following (individually and in combination), but none
> of them have solved the problem:
> 1. mpiexec -ssi boot_rsh_ignore_stderr 1 (within simulator_script)
> 2. mpiexec -ssi rsh_agent "ssh -X" (within simulator_script)
> 3. mpiexec -ssi rsh_agent "ssh" (within simulator_script)
> 4. LAMRSH = "ssh -X" export LAMRSH (before calling DAKOTA but after
> lamboot and before lamboot)
> 5. LAMRSH = "ssh" export LAMRSH (before calling DAKOTA but after
> lamboot and before lamboot)
> 6. lamboot -ssi boot_rsh_ignore_stderr 1
> 7. lamboot -ssi boot rsh -ssi boot_rsh_ignore_stderr 1
> 8. mpiexec -ssi boot rsh -ssi boot_rsh_ignore_stderr 1 (within
> simulator_script)
>
> How can I change the ssh script that LAM/MPI runs when it tries to
> return to the head node?
>
> I've attached laminfo along with the complete Error message I receive.
> I've also attached my simulator_script which is where I call mpiexec
> and run_Gamesv1p2.c which sets up the MPI environment.
>
> Thanks,
>
> Jennifer Thorne
|
