SSH Tunnel Setup (Advanced)

On KAYA

Create a slurm file to open a port on the compute node.

File: tunnel.slurm

#!/bin/bash -l
#SBATCH --job-name=tunnel
#SBATCH --partition=gpu
#SBATCH --ntasks=1
#SBATCH --cpus-per-task=16
#SBATCH --mem=16G
#SBATCH --gres=gpu:p100:1
#SBATCH --export=ALL
#SBATCH --mail-type=ALL
#SBATCH [email protected]
#SBATCH --time=3-00:00:00

# module load Anaconda3/2024.06 cuda/12.4

module list

nvcc --version

nvidia-smi

/usr/sbin/sshd -D -p 2222 -f /dev/null -h ${HOME}/.ssh/id_ecdsa

Add module load commands to .bashrc file as each Jupyter cell will run a new bash session so modules need to be loaded again and again.
```
module load Anaconda3/2024.06 cuda/12.4
```
Check if KAYA has id_ecdsa file in ~/.ssh/ directory. If not, generate one using ssh-keygen command.
```
ssh-keygen -t ecdsa -b 521 -f ${HOME}/.ssh/id_ecdsa
```

Add public key to the ~/.ssh/authorized_keys file.

cat ~/.ssh/id_ecdsa.pub >> ~/.ssh/authorized_keys

Change the permissions of .ssh.

chmod 700 ~/.ssh
chmod 600 ~/.ssh/authorized_keys

Submit the slurm file.
```
sbatch tunnel.sh
```

On Local Machine

Open your ssh config file and add the following lines.

File: ~/.ssh/config

Host login-node
  HostName kaya.hpc.uwa.edu.au
  User nchoong
  PreferredAuthentications publickey
  IdentityFile "~/.ssh/id_rsa"
  ServerAliveInterval 240
  ServerAliveCountMax 2

Host compute-node
    HostName localhost
    Port 2222
    User nchoong
    IdentityFile "~/.ssh/id_rsa"

Check if you have id_rsa file in ~/.ssh/ directory. If not, generate one using ssh-keygen command.
```
ssh-keygen -t rsa -b 4096 -f ~/.ssh/id_rsa
```

Add public key to KAYA.

ssh-copy-id -i ~/.ssh/id_rsa.pub [email protected]

Change the permissions of the .ssh.

chmod 700 ~/.ssh
chmod 600 ~/.ssh/authorized_keys

Create a bash script to open a tunnel to the compute node.

File: connect_compute_node.sh

#!/bin/bash

# Kill any existing SSH tunnel
lsof -ti:2222 | xargs kill -9
ps aux | grep "ssh -L 2222" | grep -v grep | awk '{print $2}' | xargs kill

# Get the compute node name dynamically
compute_node=$(ssh login-node squeue --me --name=tunnel --states=R -h -O NodeList | xargs)

# Check if the compute node was found
if [ -z "$compute_node" ]; then
  echo "No compute node found for the current job."
  exit 1
fi

echo "Connecting to compute node: $compute_node"

# Establish SSH tunnel
ssh -L 2222:"$compute_node":2222 login-node

# add -fN to the ssh command to run it in the background without shell
# ssh -L 2222:"$compute_node":2222 login-node -fN

Change the permissions of the script.
```
chmod +x connect_compute_node.sh
```

Or you can create a function in your .bashrc file.

function connect_compute_node() {
  # Kill any existing SSH tunnel
  lsof -ti:2222 | xargs kill -9
  ps aux | grep "ssh -L 2222" | grep -v grep | awk '{print $2}' | xargs kill

  # Get the compute node name dynamically
  compute_node=$(ssh login-node squeue --me --name=tunnel --states=R -h -O NodeList | xargs)

  # Check if the compute node was found
  if [ -z "$compute_node" ]; then
    echo "No compute node found for the current job."
    return 1
  fi

  echo "Connecting to compute node: $compute_node"

  # Establish SSH tunnel
  ssh -L 2222:"$compute_node":2222 login-node
}

Run the script or function.

./connect_compute_node.sh

connect_compute_node

Then open VSCode or another terminal and run the following command to connect to the compute node.
```
ssh compute-node
```
or
```
ssh -p 2222 nchoong@localhost
```