4. Running in Jupyter Notebook on a HPC platform

Note

This example is based on CSD3 for people associated with the University of Cambridge but the workflow can be applied to other HPC clusters.

This guide provides step-by-step instructions on how to set up a virtual environment for Jupyter Notebook on an HPC system, run Jupyter Notebook on both login and compute nodes, install the detectree2 package, and run it on a GPU compute node.

4.1. Setting Up a Virtual Environment for Jupyter Notebook

First, create a virtual environment to run Jupyter Notebook. This step only needs to be done once.

# Load the Python module
module load python/3.8

# Create a virtual environment with system site packages
virtualenv --system-site-packages ~/jupyter-env

# Activate the virtual environment
source ~/jupyter-env/bin/activate

# Install Jupyter Notebook
pip install jupyter

4.2. Running Jupyter Notebook on the Login Node (CPU Only)

Follow these steps to run Jupyter Notebook on the login node using CPU resources only.

Steps: 1. Activate the Virtual Environment:

source ~/jupyter-env/bin/activate

  1. Start Jupyter Notebook:

    jupyter notebook --no-browser --ip=127.0.0.1 --port=8081

    • This command will start a Jupyter Notebook server and output a URL that looks like http://127.0.0.1:8081/?token=.... Save this URL for later use.

    • If the 127.0.0.1 URL doesn’t work, try using the one that starts with http://localhost.

  2. Set Up SSH Tunneling from Your Local Computer:

    Open a terminal on your local computer and run:

    ssh -L 8081:127.0.0.1:8081 -N <CRSid>@<login_hostname>.hpc.cam.ac.uk

    • Replace <CRSid> with your user ID and <login_hostname> with the hostname of the login node (e.g., login-p-1).

    • Example:

      ssh -L 8081:127.0.0.1:8081 -N ab123@login-p-1.hpc.cam.ac.uk

    • Keep this terminal open as long as you want the session to last.

  3. Access Jupyter Notebook:

    • Paste the URL you saved earlier into your web browser to access the Jupyter Notebook interface.

4.3. Running Jupyter Notebook on a Compute Node (CPU or GPU)

To run Jupyter Notebook on a compute node, follow these steps.

Steps: 1. Note the Login Node Hostname:

hostname

  • Run this command on the login node and note the hostname (e.g., login-p-1).

  1. Request an Interactive Session:

    • For GPU Compute Node:

      sintr -A COOMES-SL3-GPU -p ampere -N1 -n1 --gres=gpu:1 -t 1:0:0 --qos=INTR

    • For CPU Compute Node:

      sintr -A COOMES-SL3-CPU -p icelake -N1 -n38 -t 1:0:0 --qos=INTR

      • Note that COOMES-SL3-* should be replaced with your or your lab’s balance account.

  2. Note the Compute Node Hostname:

    hostname

    • Run this command on the compute node and note the hostname (e.g., gpu-q-3).

  3. Activate the Virtual Environment:

    source ~/jupyter-env/bin/activate

  4. Load Necessary Modules (if required):

    • For example, to load CUDA:

      module load cuda

  5. Start Jupyter Notebook:

    jupyter notebook --no-browser --ip=* --port=8081

    • This will output a URL starting with http://127.0.0.1:8081/?token=.... Save this URL for later.

  6. Set Up SSH Tunneling from Your Local Computer:

    Open a terminal on your local computer and run:

    ssh -L 8081:<compute_hostname>:8081 -N <CRSid>@<login_hostname>.hpc.cam.ac.uk

    • Replace <compute_hostname> with the hostname from step 3 and <login_hostname> from step 1.

    • Example:

      ssh -L 8081:gpu-q-3:8081 -N ab123@login-p-1.hpc.cam.ac.uk

    • Keep this terminal open as long as you want the session to last.

  7. Access Jupyter Notebook:

    • Paste the URL you saved earlier into your web browser to access the Jupyter Notebook interface.

Note

If you are connected to the cluster via VSCode, it will try to forward the port automatically, which does not work. In that case you might need to shuffle ports around and connect to the new port, like: ssh -L 8082:gpu-q-3:8081 ... and then enter the link in your browser by replacing 8081 with 8082.

4.4. Installing Detectree2

Follow these steps to install the detectree2 package on a GPU compute node.

Steps: 1. Request a GPU Interactive Session:

sintr -A COOMES-SL3-GPU -p ampere -N1 -n1 --gres=gpu:1 -t 0:30:0 --qos=INTR

  1. Load Required Modules:

    module load gcc
module load cuda/12.1
module load cudnn/8.9_cuda-12.1
module load gdal/3.7.0-icl

  2. Activate the Virtual Environment:

    source ~/jupyter-env/bin/activate

  3. Install PyTorch with CUDA Support:

    pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121

  4. Install Additional Python Packages:

    pip install opencv-python
pip install GDAL==3.7.0

  5. Set CUDA Environment Variables:

    export CUDA_HOME=/usr/local/software/cuda/12.1
export PATH=/usr/local/software/cuda/12.1/bin:/usr/local/software/cuda/12.1/cuda-samples-12.1/bin/x86_64/linux/release:/usr/local/software/cuda/12.1/nvvm/bin:/usr/local/software/cuda/12.1/libnvvp:$PATH
export CPATH=/usr/local/software/cuda/12.1/include:$CPATH
export FPATH=/usr/local/software/cuda/12.1/include:$FPATH
export LIBRARY_PATH=/usr/local/software/cuda/12.1/lib64:/usr/local/software/cuda/12.1/lib:/usr/local/software/cuda/12.1/nvvm/lib64:$LIBRARY_PATH
export LD_LIBRARY_PATH=/usr/local/software/cuda/12.1/lib64:/usr/local/software/cuda/12.1/lib:/usr/local/software/cuda/12.1/nvvm/lib64:$LD_LIBRARY_PATH

  6. Install Detectree2:

    pip install git+https://github.com/PatBall1/detectree2.git

4.5. Running Detectree2 on a GPU Compute Node

When running any detectree2 tasks on a GPU compute node, ensure the following modules are loaded and the Python environment is active.

Steps: 1. Request a GPU Interactive Session:

sintr -A COOMES-SL3-GPU -p ampere -N1 -n1 --gres=gpu:1 -t 1:0:0 --qos=INTR

  1. Load Required Modules:

    module load cuda/12.1
module load cudnn/8.9_cuda-12.1
module load gdal/3.7.0-icl

  2. Activate the Virtual Environment:

    source ~/jupyter-env/bin/activate

  3. Run Your Detectree2 Tasks:

    • You can now run your Python scripts or Jupyter Notebooks that utilize detectree2.

Note

Always ensure that you’re working within the allocated time for interactive sessions and that you comply with the HPC usage policies.