[gmx-users] REMD: mdrun_mpi crash with segmentation fault (but mpi is working)

[Felipe Villanelo]

Hi,

I trying to learn REMD following the tutorial on gromacs page
<http://www.gromacs.org/Documentation/Tutorials/GROMACS_USA_Workshop_and_Conference_2013/An_introduction_to_replica_exchange_simulations%3A_Mark_Abraham,_Session_1B>
on
a 4-cores computer.
However when I try to use the command:
mpirun -np 4 mdrun_mpi -v -multidir equil[0123] (as the tutorial says)
the program crashed with the following error:
mpirun noticed that process rank 2 with PID 13013 on node debian exited on
signal 11 (Segmentation fault).

The mpi is running fine with the 4 cores if I run a simple gromacs
simulation (NPT equil) in the same machine.
So I think it is not a problem of mpi configuration (as I read in another
thread)

These with gromacs version is 5.0.2

If I try to run the same with an older version of gromacs (4.5.5) the error
is different (previously adjusting the options on the mdp file to match
changes in syntaxis betweeen versions):

[debian:23526] *** An error occurred in MPI_comm_size
[debian:23526] *** on communicator MPI_COMM_WORLD
[debian:23526] *** MPI_ERR_COMM: invalid communicator
[debian:23526] *** MPI_ERRORS_ARE_FATAL (your MPI job will now abort)

But this version also work fine with mpi using the 4 cores on a simple
simulation

Thanks
Bye

Felipe Villanelo Lizana
Bioquímico
Laboratorio de Biología Estructural y Molecular
Universidad de Chile