[lammps-users] compile REAXFF in Parallel version?

Dear all:

I recently complie lammps with REAXFF package on turing cluster. I can successfully complie it, but when I run it, it comes out a error

Dear all:

dear vivian,

I recently complie lammps with REAXFF package on turing cluster. I can
successfully complie it, but when I run it, it comes out a error
---------------------------------------------------------------------------------------------------
> Running 'lmp_crockett_2June2009' on 16 processors/8 nodes:

> Running sequential program:
/turing/home/hzhao4/REFF/lmp_crockett_2June2009
<MPICH-MX> Error: Need to obtain the job magic number in MXMPI_MAGIC !
[0] Error: write to socket failed !

this looks like you compiled for using a myrinet MX type hardware,
but the node you are running on actually does not have that kind
of hardware installed, or you have no (unix) permission to access it.

try some simple MPI test code, and you will probably see the same.

cheers,
    axel.

Dear Axel:

Thank you for your quick reply. Before REAXFF package came out, I have run LAMMPS on that turing cluster with similar MPI setup and it worked well. But after I include the reaxff package and compile lammps with the same MPI compiler, it can be compiled but it can not work properly as I showed the error in the previous email.

I am wondering is it because the reaxff lib has some specific requirements about the parallel computing setup?

Thank you~

Bests
HJ

Dear Axel:

Thank you for your quick reply. Before REAXFF package came out, I
have run LAMMPS on that turing cluster with similar MPI setup and it
worked well. But after I include the reaxff package and compile lammps
with the same MPI compiler, it can be compiled but it can not work
properly as I showed the error in the previous email.

dear vivian,

no. this message comes from deep within the bellies of your
MPI library (actually the lowlevel library it is using).
there must be a difference in what you did, or somebody
installed a different MPI library or changed the configuration
of paths or you have a node with a broken hardware (or a crashed
firmware), so something else.

I am wondering is it because the reaxff lib has some specific
requirements about the parallel computing setup?

not at this low level.

cheers,
   axel.