[gmx-users] problem running gromacs v3.3 on SGE cluster?

[Lubos Vrbka]

Cameron Mura wrote:
> Hi,
> 
> I'm wondering if anyone's experiences troubles running the latest 
> gromacs v3.3 on linux clusters using the SGE queueing system?  In a 
> nutshell -- I built gromacs against MPI libs using the intel compiler as 
> per usual procedure, and jobs which I submit for 2 or 4 procs (so "-pe 
> mpi 4" in the SGE preamble) seem to run fine... However, it fails in a 
> weird way for greater than 6 procs. Please see the emails below for more 
> detailed info about all this.  I'd greatly appreciate any advice if 
> anyone's seen anything similar (or any links to running gromacs on sge, 
> other sources of info...)
i had similar problems with gromacs (search archives for my recent post 
on parallel gromacs) - up to 4 processors it is ok. if more processors 
are used, then the mdrun processes are on the respective nodes, but do 
not do anything, except for the last one.

i used PBSPro and MPICH. this thing is probably not queueing system 
related. i was told that gromacs is knows not to work well with mpich... 
lam was recommended. i built lam, rebuilt gromacs and it seems to be ok 
(however, i didn't have time enough to run thorough tests)

-- 
Lubos
_ at _"