The aim of this recipes is learn how to use the execution queue on Ogbon Environment. A practical example to see how it can be used and to see a real example of submit jobs. The results are impressive for the effort and performance on the supercomputacional environment.

~$ ssh -p 5001 [email protected]

Please refer to create a alias for connect with OGBON, see [Simplifying SSH] for an way to reduce the complexity of this command.

By following the steps below you will be able to do just

~$ ssh ogbon and successfully connect to the server.

Create or edit a ~/.ssh/config file

~$ mkdir -p ~/.ssh

then create or edit the ~/.ssh/config file, appending the following content:

Host ogbon
    HostName ogbon-login8.fieb.org.br
    User murilo
    PreferredAuthentications publickey
    Compression yes
    ServerAliveInterval 40
    ForwardX11 yes
    Port 5001
    IdentityFile ~/.ssh/id_rsa

where you should change the name of the User option from murilo to your username. Furthermore, check if your ssh key is really id_rsa, otherwise change it to the correct one.

~$ ssh ogbon

~$ sinfo

~$ squeue

~$ module avail

~$ module load gcc/11.1.0

~$ module list

~$ groups ~$ murilo nec projetos cenpes-lde

To allocate run:

~$ salloc -p cpulongb -N 1 -A cenpes-lde

The expected output is something like:

salloc: Pending job allocation 528705
salloc: job 528705 queued and waiting for resources
salloc: job 528705 has been allocated resources
salloc: Granted job allocation 528705
salloc: Waiting for resource configuration
salloc: Nodes c153 are ready for job

With the node c153 (only an example) properly allocated, ssh into it with the following command:

~$ ssh c153

To free up the allocated resources run:

~$ scancel -u murilo

~$ sbatch script-slurm.sh

!/bin/sh

#SBATCH --job-name=MPI                         # Job name
#SBATCH --nodes=2                              # Run all processes on 2 nodes
#SBATCH --partition=cpulongb                   # Partition OGBON
#SBATCH --output=out_%j.log                    # Standard output and error log
#SBATCH --ntasks-per-node=1                    # 1 job per node
#SBATCH --account=cenpes-lde                   # Account of the group

~$ scp -P 5001 -r /Users/muriloboratto/Documents/github/howto-ogbon/ [email protected]:/home/murilo/

~$ scp -P 5001 -r [email protected]:/home/murilo/cap-hpc/ .

Click your profile photo in GitHub > Settings > SSH and GPG keys > Add SSH key

~$ ssh -p 5001 -CXY -o ServerAliveInterval=40 [email protected] -L 8559:*:8559

~$ module load anaconda3/2020.07

~$ jupyter lab --port=8559

~$ ssh -p 5001 -CXY -o ServerAliveInterval=40 [email protected] -L 8559:*:8559

To allocate run:

~$ salloc -p gpulongb -N 1 -A cenpes-lde

The expected output is something like:

salloc: Pending job allocation 528705
salloc: job 528705 queued and waiting for resources
salloc: job 528705 has been allocated resources
salloc: Granted job allocation 528705
salloc: Waiting for resource configuration
salloc: Nodes c003 are ready for job

With the node c003 (only an example) properly allocated, ssh into it with the following command:

~$ ssh c003 -L 8559:*:8559

~$ module load anaconda3/2020.07

~$ jupyter lab --port=8559

~$ https://hub.docker.com/search?q=speglich

~$ singularity pull docker://speglich/cimatec-base

~$ singularity exec --nv docker://speglich/cimatec-base bash

Then, read the instructions that follow in the notebook, and connect with NICE DCV on OGBON:

~$ ssh -p 5001 [email protected]

~$ alias dcvCreate="dcv create-session profiling"

~$ alias dcvList="dcv list-sessions"

~$ alias dcvClose="dcv close-session profiling"

~$ dcvCreate

https://ogbon-cgpu4.fieb.org.br:8443#profiling

After clicking the Connect button you will be asked for a password, which is registered in the NOC/CS2I.

~$ module load anaconda3/2023.07

~$ conda info --envs

[murilo@login8 ~]$ conda info --envs
# conda environments:
#
pytorch-2.x              /home/murilo/.conda/envs/pytorch-2.x
tensorflow-2.x           /home/murilo/.conda/envs/tensorflow-2.x
base                  *  /opt/share/anaconda3/2020.07
llvm12                   /opt/share/anaconda3/2020.07/envs/llvm12

~$ source activate pytorch-2.x

1. Create a reference file called conda-pytorch-env.yaml:

name: pytorch-2.x
channels:
  - pytorch
  - conda-forge
  - nvidia
dependencies:
  - python=3.11
  # Libraries
  - pytorch-cuda=11.8
  - pytorch>=2.0.1
  - numpy
  - pandas
  # Tools
  - ipykernel
  - jupyterlab
  - pip

2. Create the env:

~$ conda env create --name pytorch-2.x --file conda-pytorch-env.yaml

3. And activate:

~$ source activate pytorch-2.x

~$ source deactivate

[murilo@login8 ~]$ ls /public/singularity/tensorflow-2.14-gpu-jupyter.sif

[murilo@login8 ~]$ sbatch /public/singularity/slurm-jupyter-notebook.sh

    ssh -L 9807:c000:8888 [email protected] -p 5001

 http://127.0.0.1:9807/tree?token=f79af719ee03701f0a0cf2f02cc72e8a895800487d595559