I am running into a problem on two of the Linux clusters I use for running Star-CCM+ parallel simulation.

On both the clusters after running for some time correctly somehow the parallel communication will freeze on one of the nodes and literally kill it and hence the parallel server will freeze. Only way to get out is Server -> Kill. Then when I restart it starts from last 'Auto Save' but there are times by doing this the simulation does get corrupted and hence I can't restart the simulation and have to start from scratch.

Anyone else having this kind of problem? If so, is there a fix?

Thanks in advance,

SG

What version of the code are you using and with which MPI?

These are the mpich on the system

lrwxrwxrwx 1 root root 11 Feb 13 2006 mpich -> mpich-1.2.4 drwxr-xr-x 12 root root 288 Oct 29 2002 mpich-1.2.1 drwxr-xr-x 12 root root 288 Oct 29 2002 mpich-1.2.2.3 drwxr-xr-x 12 root root 288 May 24 2002 mpich-1.2.4 lrwxrwxrwx 1 root root 5 Feb 13 2006 mpich-ch_p4 -> mpich

problem is happening with versions 2.10.013 as well as version 3.

You try should use HP-MPI that is included with STAR-CCM+, not mpich.

Actually one of the suggestions directly from Adapco was to use mpich

" If not, the default MPI communication is HPMPI2. Try using the option and specify "-mpidriver mpich" to use mpich "

So I have been using the default mpi so should be the HPMPI. So I think the problem is not in the MPI communication code BUT more with the solver's code.

My sysadmin is looking at architecture as we speak to resolve this issue BUT I thought if someone else had faced a similar problem then the fix could be communicated faster to him.

How much memory do you have on each node and what kind of interconnect (GigE, Infiniband,...)?