::Centre for Theoretical Studies::

Tutorials

Accessing the cluster

To access cluster you need to “ssh” the login server.
The master node is primary gateway to the rest of the cluster, which has a job scheduler (called SGE). You may submit jobs to the queue and they will run when the required resources are available.
Please do not run programs directly on login node. Login node is use to submit jobs, transfer data and to compile source code.
By default, user data save in /home directory. These directory are available on master node as well as the other nodes on the cluster.

Remote access from Windows Computer

Using any open source “ssh” client applications for Windows you can login and access the cluster. Enter the user name along with IP address or Hostname with which you wish to connect.
e.g. [ username @ 10.33.1.18 / username @ 10.33.1.59 ]
Enter your password when prompted, and press Enter.

Remote access from Linux Computer

Linux systems provide a built-in SSH client, so there is no need to install any additional package. Open the terminal, connect to an SSH server by typing the following command:
ssh [username]@[hostname]

SGE Job Control System

In order to submit a job, one needs to have a script that contains all the information regarding the job. A sample script is shown below. Once you are ready with your script (e.g. myjob.sh), you can submit and manage your job using the following commands.

To submit a job : qsub myjob.sh
To check the status of your jobs : qstat
To check availability of the nodes : qstat -f
To delete/cancel a submitted job : qdel
To get the names of the queues : qconf -sql

1. For serial jobs:

#!/bin/bash --login
#$ -N myjob # name of the job
#$ -q all.q # queue names can be found using " qconf -sql "
#$ -l h_rt=01:00:00 # walltime in HH:MM:SS
#$ -j y # merge the output and error in a single file
#$ -cwd # submit from current working directory
#$ -V # load the environment variables

# name of the executable
exe = /path/of/the/executable

# run the application
$exe [arguments]

2. For parallel MPI jobs:

#!/bin/bash
#$ -N myjob # name of the job
#$ -q all.q # queue names can be found using " qconf -sql "
#$ -pe mpi 4 # no of processes or tasks
#$ -l h_rt=01:00:00 # walltime in HH:MM:SS
#$ -j y # merge the output and error in a single file
#$ -cwd # submit from current working directory
#$ -V # load the environment variables

# name of the executable
exe = /path/of/the/executable

# run the application
mpirun -np $NSLOTS $exe [arguments]

3. For parallel openMP jobs:

#!/bin/bash
#$ -N myjob # name of the job
#$ -q all.q # queue names can be found using " qconf -sql "
#$ -pe smp 4 # no of processes or tasks
#$ -l h_rt=01:00:00 # walltime in HH:MM:SS
#$ -j y # merge the output and error in a single file
#$ -cwd # submit from current working directory
#$ -V # load the environment variables

export OMP_NUM_THREADS=$NSLOTS

# name of the executable
exe = /path/of/the/executable

# run the application
$exe [arguments]

Examples

Running VASP using Intel libraries:

#!/bin/bash
#$ -N vasp_job # name of the job
#$ -q all.q # queue names can be found using " qconf -sql "
#$ -pe mpi 4 # no of processes or tasks
#$ -l h_rt=01:00:00 # walltime in HH:MM:SS
#$ -j y # merge the output and error in a single file
#$ -cwd # submit from current working directory
#$ -V # load the environment variables

# load intel compilers and libraries
source /opt/intel/bin/compilervars.sh intel64

# name of the executable
exe = vasp

# run the application
mpirun -np $NSLOTS $exe

Note:

Always try to use the full path of your executable.
Verify the path of the intel library before using this script.