On Mon, Jan 28, 2013 at 2:16 AM, Jorgen Simonsen <jorgen589.gmail.com>wrote:
> Hi all,
>
> I am trying to install a mpi-version of amber10 on our cluster. I have
> just installed the serial version and it works fine.
>
> module list gives me this
>
> 1) modules 2) intel-mpi
>
> and I configure with the following options
>
> ./configure_amber -intelmpi -nobintraj -verbose -nosanderidc gfortran
>
> which gives me this
>
> Installation of Amber10 (parallel) is complete at Sun Jan 27 23:09:42 PST
> 2013.
>
> next I go to ../test and
>
> export DO_PARALLEL='mpirun -np 4'
>
> and make test.parallel which gives me this error
>
> cd cytosine && ./Run.cytosine
> /etc/tmi.conf: No such file or directory
> /etc/tmi.conf: No such file or directory
> /etc/tmi.conf: No such file or directory
> /etc/tmi.conf: No such file or directory
> [1] MPI startup(): tmi fabric is not available and fallback fabric is
> not enabled
> [2] MPI startup(): tmi fabric is not available and fallback fabric is
> not enabled
> [0] MPI startup(): tmi fabric is not available and fallback fabric is
> not enabled
> [3] MPI startup(): tmi fabric is not available and fallback fabric is
> not enabled
>
> and on one of the nodes I get the following error
> sander.MPI: error while loading shared libraries: libmpi.so.4: cannot
> open shared object file: No such file or directory
>
> how to fix this? Thanks
>
Sounds like your MPI is not properly set up. Is there an intel compiler
module? (module avail intel) If so, try loading the same intel compiler as
the intel MPI you have loaded.
Another issue may be that the shell containing your job did not load those
modules. Especially if you are using PBS, you cannot assume the
environment with which you submitted the job will the the environment of
the job when it begins running. A notable difference is that the shell you
submit from is interactive (i.e., PS1 is defined), and a job you submit is
not (unless you submit with -I). Many system *rc files have different
setups for interactive vs. non-interactive shells.
Running the command "module list" before running sander.MPI should dump the
active modules to the PBS output file so you can take a look.
You could also try a different MPI. I've never used the Intel MPI,
personally, and I don't know if it offers much advantage over MPICH2 (or
MPICH, as it's called these days).
HTH,
Jason
--
Jason M. Swails
Quantum Theory Project,
University of Florida
Ph.D. Candidate
352-392-4032
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Mon Jan 28 2013 - 05:00:02 PST
