[gmx-users] mpich parallel system errors
Mark.Abraham at anu.edu.au
Sat Dec 17 14:26:42 CET 2005
Rongliang Wu wrote:
> Hello gmx-users,
> i substrated a 5 nodes job to a mpich cluster, but i found three of the nodes are not working.
> the system manager told me that there are no problems with the hardware.
If there had been, gromacs would probably have failed with an MPI error.
> can it be any problem with my input files?
Yes it can be, particularly if you'd constructed and run a 2-node job :-)
> the job ended at end with the two nodes, will the results from just two nodes worth trust?
We can't tell anything from what you've told us.
One non-definitive test would be to take your final geometry and do a
short run from that point and see if the final energy components from
the first run are the same as the initial energy components from the
subsequent short run.
More information about the gromacs.org_gmx-users