[gmx-users] Gromacs 4.0.2 hangs?

[Yves Dubief]

Dear all,

I have some difficulties running mdrun_mpi for large number of  
iterations, typically > 40000. The code hangs, not necessarily at a  
fixed iteration, for all configurations I run  and this behavior shows  
up under linux and mac OS. The simulations range from 10^4 to 10^6  
coarse-grained atoms and I try to keep the amount of data generated  
per much smaller than 1Gb. I have worked around this issue by running  
25000-iteration similations with tpbconv-restarts, as follows:
mpiexec -np 256 mdrun_mpi -nosum -v -s dppc0_1.tpr -o dppc0_1.trr -c  
dppc0_1.gro -e dppc0_1.edr -x dppc0_1.xtc
  tpbconv -s dppc0_1.tpr -f dppc0_1.trr -e dppc0_1.edr -o dppc0_2.tpr - 
extend 250.0
With such scripts, I have been able to generate 10^5 to 10^6  
iterations without any problem. I was wondering if anyone has  
experienced similar problems and if I am missing something.

I have pretty much ruled out a problem with mpi, since I have  
thoroughly tested these computers  with other mpi codes. I am now  
wondering if there might be a problem with output files.

I run Gromacs 4.0.2 on a linux cluster (quad core processors, myrinet  
and mpich, compiled with gcc, single precision) up to 256 processors,  
on a macpro 2 quad  and on a macbook dual core using the Fink package  
for openmpi. All these computers have enough available disk space for  
any simulation I run. A typical simulation is a coarse grained MD  
using martini with .mdp files obtained from Marrink's website.

Best,

Yves



--
Yves Dubief, Ph.D., Assistant Professor
Graduate program coordinator
University of Vermont, School of Engineering
Mechanical Engineering Program
201 D Votey Bldg, 33 Colchester Ave, Burlington, VT 05405
Tel: (1) 802 656 1930 Fax: (1) 802 656 3358
Also:
Vermont Advanced Computing Center
206 Farrell Hall, 210 Colchester Ave, Burlington, VT 05405
Tel: (1) 802 656 9830 Fax: (1) 802 656 9892
email: ydubief at uvm.edu
web: http://www.uvm.edu/~ydubief/