swagat mishra wrote: > hello everyone, > i have a network of systems connected over lan with each computer > running ubuntu. openmpi 1.4.x is installed on 1 machine and the > installation is mounted on other nodes through Networking File > System(NFS). the source program and compiled file(a.out) are present in > the mounted directory > i run my programs by the following command: > /opt/project/bin/mpirun -np 4 --prefix /opt/project/ --hostfile > hostfile a.out > i have not set LD_LIBRARY_PATH but as i use --prefix mpirun works > successfully > > however as per the open mpi debugging faq: > http://www.open-mpi.org/faq/?category=debugging > when i run > /opt/project/bin/mpirun -np 4 --prefix /opt/project/ --hostfile > hostfile -x DISPLAY=10.0.0.1:0.0 xterm -e gdb a.out > > 4 xterm windows are opened with gdb running as expected. however when i > give the command start to gdb in the windows corresponding to remote > nodes, i get the error: > libmpi.so.0 not found: no such file/directory > > as mentioned other mpi jobs run fine with mpirun > > when i execute > /opt/project/bin/mpirun -np 4 --prefix /opt/project/ -x > DISPLAY=10.0.0.1:0.0 xterm -e gdb a.out ,the debugging continues succesfully > > please help >
You need to set LD_LIBRARY_PATH to include the path to the OpenMPI libraries. The --prefix option works for OpenMPI only; it has no effect on other programs. You also need to make sure that the LD_LIBRARY_PATH variable is correctly passed along to the other OpenMPI programs. For processes on other hosts, this is usually done by editing your shell's rc file for non-interactive logins (.bash_profile for bash). -- Prentice