|
Acces to cluster Access to the Ibnqonfud Cluster of the University of Skikda is via an SSH connection :
• Under windows: use Teraterm ou putty • Under linux: use openssh Access with command line via ssh for a user with user ,ame user is:
ssh
Compilation of programs Default compilators on Ibnqonfud are of Intel : icc et ifort.
You can compile your programs (fortran or C) using following commands : • On fortran : mpiifort your_program.for -o your_executable • On C : mpiicc your_program.c -o votre_executable It is highly recommended to test your executable's dependency on math libraries by the command: Idd votre_executable
Submission of jobs (tasks)
Le cluster Ibnqonfud uses scheduler slurm .In this effect, it is recomanded to write a script bash (or c-shell).
Example :
#!/bin/bash #SBATCH --partition=R424# Name of partition : R424 #SBATCH –nodes=5 # Number of used nodes 5 #SBATCH –ntasks-per-node=20# Number of cores by node : 20(maximum) #SBATCH –time=00:10:00 # Time by Day-Hour:minutes:secondes #SBATCH –mail-type=END, FAIL # Notification by mail of job #SBATCH –mail-user=your_mail # mail address of user #SBATCH –error=ibnqonfud.%N.%j.err # Standard out for errors #SBATCH –output=ibnqonfud.%N.%j.out # Standard out for results srun ./Pi_mpi.x Notice: A job can be sequential, parametric, or parallel, for documentation, go to this site Example1.Calculation of Pi using the Monte Carlo method : (look for Source code ) 2. Compile programme with: $ mpiicc -O2 Pi_mpi.c -o Pi_mpi.x 3. Write a batch file Pi_mpi.sh pour le lancer dans le cluster : The content of file Pi_mpi.sh est:
#!/bin/bash
#SBATCH --partition=r424 #SBATCH --nodes=2 #SBATCH --ntasks-per-node=20 #SBATCH -J Pi_mpi #SBATCH –time=0-00:10:00 #SBATCH --mail-type=end #SBATCH –mail-user= #SBATCH --error=Pi_mpi.err #SBATCH --output=Pi_mpi.out srun ./Pi_mpi.x 4. Submit script to Slurm : $ sbatch Pi_mpi.sh Calculations follow-up
• To check its calculations, it is necessary to execute:
squeue -u • For cancelling a submission,you have to identify the job ID with squeue and execute, scancel • To view detailed job information : scontrol show job jobid -dd • To estimate the likely start time of a waiting job : scontrol show job | grep StartTime= ou squeue -o "%S" -j |