[gmx-users] md using mpich

[gianluca santarossa]

Hi all,
    I 'm in trouble when launching molecular dynamic simulations in 
parallel with more then 4 procs.
First of all, I can launch minimizations using any number of procs and 
it ends correctly.
Moreover, when I run the simulation with 4 or less procs, everything 
works fine.
But if I launch the same simulation using 5 procs or more it doesn't 
work. The odd thing is that it doesn't exit with errors, but it waits 
for data which never arrive.

I'm using gromacs 3.2 compiled from source on a cluster mounting Rocks 
(Red-hat based).
My command is just

/opt/mpich/gnu/bin/mpirun -np 6 -nolocal -machinefile machines 
/usr/local/bin/mdrun_mpi -np 6 -v

and the output is something like

[cut]
starting mdrun 'Generated by genbox'
100 steps,      0.2 ps.


and, after some time, it writes out
p4_1636:  p4_error: Timeout in establishing connection to remote process: 0
Received disconnect from 192.168.1.250: Command terminated on signal 13.


I repeat: other software run, and even the minimization with gromacs 
works fine.
I don't know what's happening. Any ideas?

Thanks
Gianluca