2.4 Installation on Cluster

2.4.1 Compile MOKIT on Cluster(集群)

If you already install MOKIT using conda, or you prefer using pre-built packages of MOKIT, you can skip this subsection. This is for compiling MOKIT from the source code on a Cluster. If you can write environment variables freely into ~/.bashrc, then there is nothing different with installation on your local Linux computer.

However, sometimes you are not allowed, or you do not want to modify ~/.bashrc. And usually environment variables of various packages (e.g. ifort, python, etc) are controlled by the module function. For example, if you need some version of python (other than system default), you have to run module load python/... to make it accessible. In such case, you have to firstly load proper versions of Intel compiler and Python before installing MOKIT. For example,

module avai                 # find which compilers/versions you can use
module load intel/2019u5    # load the Intel version you want to use
module load anaconda/2024.2 # load the Python version you want to use
cd $MOKIT_ROOT/src          # enter MOKIT source code directory
make all                    # compile MOKIT

The first line finds which versions of software packages you can choose. The 2nd and 3rd lines load the corresponding software. The 4th line is that you enter the MOKIT source code directory. And the 5th line is compiling MOKIT. This is just a simple example telling you how to load modules on a Cluster. For detailed questions, please consult the administrator of your Cluster.

2.4.2 Install MOKIT on Cluster

For online installation, there is no difference between personal computer and Cluster, please read 2.2.1 Online Installation. If you want to compile MOKIT from source code, read Section 2.4.1 above.

2.4.3 Use MOKIT on Cluster

Usually you are not allowed to submit any job in the main node or login node. And there exists a queue/batch system on your cluster, e.g. bsub of LSF, qsub of PBS, or sbatch of SLURM, which distributes your job to a certain computation node. Taking bsub as an example, you should write a Shell script like

#!/bin/sh
#BSUB -q ... # queue name
#BSUB -n 4 # number of cores

module load intel/2019u5
module load anaconda/2024.2
export MOKIT_ROOT=$HOME/software/mokit
export PATH=$MOKIT_ROOT/bin:$PATH
export PYTHONPATH=$MOKIT_ROOT:$PYTHONPATH
export GMS=$HOME/software/gamess/rungms

gjfname=$1
outname=${gjfname%.*}.out
automr $gjfname > $outname 2>&1

Here module load and export assumes that you compiled MOKIT from its source code, using some specific compiler and Python. So corresponding libraries and environment variables are set firstly. But if you install MOKIT using conda install, the script may be slightly different

#!/bin/sh
#BSUB -q ... # queue name
#BSUB -n 4 # number of cores

source activate mokit-py39
gjfname=$1
outname=${gjfname%.*}.out
automr $gjfname > $outname 2>&1

where everything is well set in the virtual environment mokit-py39. Thus only an activation is needed. Next you can run

bsub run.sh 00-h2o_cc-pVDZ_1.5.gjf

to submit the job.

If you are using the SLURM, the Shell script would look like

#!/bin/bash
#SBATCH -J automr
#SBATCH -e automr.err.%j
#SBATCH -p cpu # queue name
#SBATCH -N 1
#SBATCH -n 1   # number of MPI processors
#SBATCH -c 32  # number of OpenMP threads
#SBATCH --mem=64G

source activate mokit-py39
gjfname=$1
outname=${gjfname%.*}.out
automr $gjfname > $outname 2>&1

Here source activate mokit-py39 assumes that the MOKIT is installed in the virtual environment mokit-py39, so we activate the target environment before using MOKIT. If you compile MOKIT from source code and use no virtual environment, there is no need to write this line. Next you can run

sbatch run.sh 00-h2o_cc-pVDZ_1.5.gjf

to submit the job. This job will call the quantum chemistry packages Gaussian, GAMESS and PySCF in succession, and all three packages are of OpenMP parallelism.

Important: source activate mokit-py39 means that activating the virtual environment (before running automr). It is not recommended to write conda activate mokit-py39. And it is recommended that you submit the SLURM script in the (base) environment. Please do not enter (mokit-py39) in advance since it is done in the script via source activate.

If some MPI programs (e.g. ORCA) would be called during computations, remember to swap the parameters of -n and -c. For example, if mokit{CASSCF_prog=ORCA} is specified in your input file, you will need a Shell script like

#!/bin/bash
#SBATCH -J automr
#SBATCH -e automr.err.%j
#SBATCH -p ... # queue name
#SBATCH -N 1
#SBATCH -n 32  # number of MPI processors
#SBATCH -c 1   # number of OpenMP threads
#SBATCH --mem=64G

source activate mokit-py39
gjfname=$1
outname=${gjfname%.*}.out
automr $gjfname > $outname 2>&1

Do not worried about -c 1. OpenMP programs can also run parallelly.

If you compile MOKIT from source code and you want to run a small MOKIT job on the current node without modifying your ~/.bashrc, you can create a Shell script, e.g. run.sh, with

module load intel/2019u5
module load anaconda/2024.2
gjfname=$1
outname=${gjfname%.*}.out
automr $gjfname > $outname 2>&1

written in this script. Then you can run

chmod +x run.sh
./run.sh 00-h2o_cc-pVDZ_1.5.gjf &

If you still encounter problems of installing or running MOKIT on your cluster, please consult the administrator of your Cluster. This is beyond the scope of this manual.