[gmx-users] cluster problem

[gerph at correo.unam.mx]

Hi everybody,

we have a cluster beowulf with smp nodes running (almost well, but not
totally) GROMACS but we can only run the demo s-peptide as just as 5
task, no more than that.
If we try to use both processors of the node it fails, even if we only
create 2 tasks and try to run them in the same node. Apparently, one of
the task hangs up (the master or task 0) and the process never ends.
I know that 'cause I kill the job taking out from pbs scheduler.

We can run that demo with one process (P) for each node;
       with P <= 5

We're using mpiexec and pbs.

Test it with gromacs 3.1.1 & 3.2.1


Next, we sumarize our cluster's configuration:


%mpiexec -v

  Version 0.71, configure options: --with-pbs=/usr/local/pbs
--with-default-comm=mpich-p4 --with-mpicc=/local/mpich/bin/mpicc


mpich-1.2.4


OpenPBS_2_3_16


All nodes configuration:         2.4.18papikernel #2 SMP  RH 7.2



Gromac's version:                3.1  &  3.2.1



Hope you can help us.


-------------------------------------------------
www.correo.unam.mx
UNAMonos Comunicándonos