multi-replica simulation

LAMMPS LAMMPS Mailing List Mirror
#1

Dear LAMMPS developers and users,

Greetings! I am curretly using 9Aug13 version on my HPC with -partition option. PBS script is as attached. However, the whole process exited as the fastest replica reaches the end (i.e., only one replica makes it to completion). However this is not seen in another HPC…so I am guessing it to be platform related. Could anyone direct me to the code piece about this so that I can try to debug? Thanks for the help and infomation

LC Liu

#!/bin/bash
#PBS -N paraffin
#PBS -o out
#PBS -e err
#PBS -q hp
#PBS -l nodes=9:ppn=8
cd $PBS_O_WORKDIR

echo ‘=======================================================’
echo Working directory is $PBS_O_WORKDIR
echo "Starting on hostname at date"

if [ -n “$PBS_NODEFILE” ]; then
if [ -f PBS_NODEFILE ]; then echo "Nodes used for this job:" cat {PBS_NODEFILE}
NPROCS=wc -l < $PBS_NODEFILE
fi
fi

mpirun -hostfile $PBS_NODEFILE -n 69 /lustre/lwork/lcliu/work/0726/lammps-9Aug13/src/lmp_c238 -p 69x1 -l spce.out -sc none -in.spce

echo "Job Ended at date"
echo ‘=======================================================’

#2

Dear LAMMPS developers and users,

Greetings! I am curretly using 9Aug13 version on my HPC with -partition
option. PBS script is as attached. However, the whole process exited as the
fastest replica reaches the end (i.e., only one replica makes it to
completion). However this is not seen in another HPC....so I am guessing it
to be platform related. Could anyone direct me to the code piece about this
so that I can try to debug? Thanks for the help and infomation

why don't you just make the following change?

diff --git a/src/main.cpp b/src/main.cpp
index 8fccaf9..fd8a4eb 100644
--- a/src/main.cpp
+++ b/src/main.cpp
@@ -31,5 +31,6 @@ int main(int argc, char **argv)
   lammps->input->file();
   delete lammps;

+ MPI_Barrier(MPI_COMM_WORLD);
   MPI_Finalize();
}

this way you _force_ all tasks to wait on the barrier before you call
finalize. if MPI_Barrier() doesn't work, you have to sue the provider
for not delivering a working MPI. :wink:

axel.

axel.

#3

Hi, Axel,

Thank you so much for providing the solution. I will try it later :slight_smile:
LC Liu

2013/7/29 下午11:35 於 “Axel Kohlmeyer” <[email protected]…33…24…> 寫道：