Make sure you have a shared-libraries LAM or OMPI installation available in all nodes in your cluster. You might use the command: $ which mpirun # discover which one comes first in your path ~/openmpi-1.2.3/bin/mpirun # might be this ~/lam-7.1.3/bin/mpirun # or this other $ ls `laminfo -path libdir | cut -d: -f2` # if using LAM lam liblamf77mpi.so.0.0.0 liblammpi++.so.0 libmpi.la liblam.a liblam.la liblammpi++.so.0.0.0 libmpi.so liblamf77mpi.a liblammpi++.a liblam.so libmpi.so.0 liblamf77mpi.la liblammpi++.la liblam.so.0 libmpi.so.0.0.0 liblamf77mpi.so liblammpio.a liblam.so.0.0.0 liblamf77mpi.so.0 liblammpi++.so libmpi.a $ ls `ompi_info -path libdir | cut -d: -f2` #if using OMPI libmca_common_sm.la libmpi_f77.so.0 libopen-pal.la libmca_common_sm.so libmpi_f77.so.0.0.0 libopen-pal.so libmca_common_sm.so.0 libmpi_f90.la libopen-pal.so.0 libmca_common_sm.so.0.0.0 libmpi_f90.so libopen-pal.so.0.0.0 libmpi_cxx.la libmpi_f90.so.0 libopen-rte.la libmpi_cxx.so libmpi_f90.so.0.0.0 libopen-rte.so libmpi_cxx.so.0 libmpi.la libopen-rte.so.0 libmpi_cxx.so.0.0.0 libmpi.so libopen-rte.so.0.0.0 libmpi_f77.la libmpi.so.0 mpi.mod libmpi_f77.so libmpi.so.0.0.0 openmpi Notice the libXXX.so dynamic libraries, as opposed to the static libXXX.a ones. Recompile LAM (configuring with --enable-shared) or OMPI (not overriding with --disable-shared the default) if you have no .so libraries. Do not forget to double-check that the same installation is available in all nodes. For instance, this should find the same libraries in all nodes (change hostnames hn to match your cluster): $ for i in h1 h2 h4 h7 h8; do ssh $i ldd `which mpirun`; done $ for i in h1 h2 h4 h7 h8; do ssh $i ldd `which mpirun` | grep lam; done $ for i in h1 h2 h4 h7 h8; do ssh $i ldd `which mpirun` | grep open; done $ for i in h1 h2 h4 h7 h8; do ssh $i ldd `which mpirun` | grep mpi; done
	Make sure your octave executable has been compiled with DLD support, and it's the same version in all cluster nodes. Type octave_config_info at the octave prompt and look for the "dld" flag, if you're not sure. If you don't have it, recompile Octave using the config switch --enable-shared. $ octave octave:1> octave_config_info ans = { dld = 1 ... DL_LDFLAGS = -shared ENABLE_DYNAMIC_LINKING = true ... SHARED_LIBS = true SHLEXT = so SHLEXT_VER = so.2.9.10 SH_LD = g++ SH_LDFLAGS = -shared SONAME_FLAGS = -Wl,-soname -Wl,oct-conf.h STATIC_LIBS = false version = 2.9.10 ... unix = 1 windows = 0 } Notice the dld=1 line. Recompile Octave (configuring with --enable-shared) if you haven't it. For instance, you could use these to check version in all nodes: $ for i in h1 h2 h4 h7 h8; do ssh $i which octave ; done $ for i in h1 h2 h4 h7 h8; do ssh $i octave -v | head -1; done
	Make sure the octave binary you wish/choose is in your search path on each node as well as the desired/chosen LAM/OMPI binaries. Also double-check that the corresponding LAM/OMPI libraries (same version / configure switches as the chosen LAM/OMPI binaries) are included in your LD_LIBRARY_PATH. That's frequently overlooked. Lots of pitfalls here: editing .profile instead of .bashrc, using double instead of single quotes in ssh commands, believing an ssh interactive session will have the same environment as a remote ssh command... Upgrading headed nodes and forgetting about slave nodes is common pitfall too. For instance, use this command, changing hostnames as appropriate: $ for i in h1 h2 h4 h7 h8; do ssh $i echo $i : '$PATH'; done h1 : /home/user/octave-2.9.10/bin:/home/user/lam-7.1.3/bin:/usr/bin:[...snip...]:/home/user/bin h2 : /home/user/octave-2.9.10/bin:/home/user/lam-7.1.3/bin:/usr/bin:[...snip...]:/home/user/bin h4 : [...snip...] $ for i in h1 h2 h4 h7 h8; do ssh $i echo $i : '$LD_LIBRARY_PATH'; done h1 : /home/user/lam-7.1.3/lib: h2 : /home/user/lam-7.1.3/lib: h4 : /home/user/lam-7.1.3/lib: h7 : /home/user/lam-7.1.3/lib: h8 : /home/user/lam-7.1.3/lib: Don't use double quotes instead of single ones around PATH    (Read bash manpage, section on QUOTING) You could also use the "env" command instead of "echo...". You can also use "which octave" or "which mpirun", just to make sure. $ for i in h1 h2 h4 h7 h8; do ssh $i echo $i : `which octave`; done h1 : alias octave=octave -q /home/user/octave-2.9.10/bin/octave h2 : alias octave=octave -q /home/user/octave-2.9.10/bin/octave h4 : alias octave=octave -q /home/user/octave-2.9.10/bin/octave h7 : alias octave=octave -q /home/user/octave-2.9.10/bin/octave h8 : alias octave=octave -q /home/user/octave-2.9.10/bin/octave
	A shared (among nodes, cluster-wide) HOME is strongly advised, as well as installing there all the required software. That way, you won't depend on anybody else. If not, you may need to play with your configuration and ask your sysadmin until the above requirements (or equivalent ones for non-ssh clusters) are met. Most common pitfalls are static-library/misconfigured LAM installs and different Octave/LAM/OMPI versions on different cluster nodes.
	Unzip & untar the toolbox. Recomended location is under ~/octave,  so a new directory ~/octave/mpitb appears. Enter the new mpitb subdir, read/edit the lam-bhost.def file to describe your cluster and lamboot your LAM (not applicable if using OMPI). Run there octave and try some MPITB command such as MPI_Initialized or MPI_Init. It might work out-of-the-box. If not, your octave version or LAM/OMPI configure switches are probably too different and MPITB needs a remake: just enter the src subdir, make sure the commands "octave-config", "mkoctfile", "(lam/ompi_)info" and "mpiCC" all work and correspond to your desired Octave / LAM/OMPI --enable-shared versions, and type "make". Read the README file for more detailed instructions. Read the Makefile to understand why those 4 commands are required. Work through the demos to learn to use the toolbox. The recommended order is Hello, Pi, PingPong, Mandelbrot, NPB/EP. Remember to halt LAM when you leave (also not much of a problem if you let the daemons there)