How to Ogbon Environment

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.

---

How to connect with ssh?

~$ 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.

Simplifying SSH

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

---

How to get information about the partitions?

~$ sinfo

---

How to get information about the SLURM queue?

~$ squeue

---

How to visualize and start the modules/APIs?

~$ module avail

~$ module load gcc/11.1.0

~$ module list

---

How to know my personal account (group of execution) on Ogbon?

~$ groups ~$ murilo nec projetos cenpes-lde

---

How to alloc/disalloc a node on Ogbon?

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

---

How to submit a job in SLURM?

~$ sbatch script-slurm.sh

bash SLURM

!/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

---

How to make scp on Ogbon?

localhost --> ogbon

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

ogbon --> localhost

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

---

It is possible make a git clone on Ogbon?

1) Only in front-end login8

2) You must associated you 'SSH keys' with your GitHub account. How?

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

---

It is possible use the jupyter lab on Ogbon in the login node?

1) Connect set the jupyter port

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

2) Initialize the Anaconda API

~$ module load anaconda3/2020.07

3) After start the jupyter lab, cut and past the link in the browser

~$ jupyter lab --port=8559

---

It is possible use the jupyter lab on Ogbon in the processing node?

1) Connect set the jupyter port

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

2) How to alloc a node on Ogbon?

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

3) On the processing node initialize the Anaconda API

~$ module load anaconda3/2020.07

4) After start the jupyter lab, cut and past the link in the browser

~$ jupyter lab --port=8559

---

How to use the Docker Containner on Ogbon?

Search the package in docker page:

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

Download the package with singularity pull:

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

Execute the package with singularity exec:

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

---

How to use NICE DCV?

Open the Remote Desktop

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

1) Connect in login8 on OGBON

~$ ssh -p 5001 [email protected]

2) Create the Alias

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

~$ alias dcvList="dcv list-sessions"

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

3) Create Session in NICE DCV

~$ dcvCreate

4) Open the browser and connect in the adress associating the alias session

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

5) Insert the user and password

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

---

How to use pytorch on Ogbon using environments?

First you have to verify all installed envs in the platform

~$ 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

If exist the env just call Pytorch with the command activate

~$ source activate pytorch-2.x

else

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

and the end, deactivate the env

~$ source deactivate

---

How to use TensorFlow with Jupyter Notebook on Ogbon using singularity image?

Singularity/Apptainer Image (From Docker Image)

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

Submit script using SLURM

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

Atention 1: Update the PATHS in the file slurm-jupyter-notebook.sh;

Atention 2: The SLURM generate a file slurm-notebook-*.log, with all informations;

Atention 3: Open a new terminal, and copy the direction select by the SLURM:

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

Atention 4: In the below of the file slurm-notebook-*.log, copy the jupyter's weblink, and change the port 8888, for the port select by SLURM, in this case 9807:

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

---