1  Getting started with HPC systems

1.1

AI-generated (Gemini Advanced 2.0). Prompt: Generate an image of a caricaturized HPC computing resources of a research institute in genomics and developmental biology. Make sure image does not include any text.

This is a short and [hopefully] simple tutorial for guiding people on how to use a HPC cluster for running their analyses. This is intended to be very generic so it is not covering a particular HPC for a specific institution. Please contact your corresponding HPC system administrator for requesting access to computing resources.

1.2 What is a HPC cluster?

A HPC cluster is the combination of:

• Many individual machines, each referred to as “nodes”

• Fast shared storage, accessible to all nodes

• All interconnected over high speed networks and/or specialized interconnects

• With resource access managed by a scheduler (i.e. slurm)

1.3 Get started with your analyses

A typical workflow on an HPC cluster includes:

• Log in: Use the command line SSH or web interface to access the cluster

  • An example on how to use the command line to log in via SSH to BSE-HPC:

    • $ ssh user@hpc.institution.edu

• Transfer Data: Move data from your local computer and/or other sources to the HPC cluster

• Find Software: Access existing software from the cluster, download from a remote source, or compile your own code

  • As an example, you can find and load existing software via module :

    # To list the available/installed software
    user@login1:~$ module avail
    --------------------- /institution/hpcdata/software/rhel9/modules/bio ---------------------------------   
    alphafold/2.3.2             bwa/0.7.17                   guppy/6.0.1           metaxa2/2.2.3            
    
    # To load software, i.e. alphafold
    user@login1:~$ module load alphafold/2.3.2
    
    # Ready to use alphfold
    user@login1:~$ alphafold
    Usage: /institution/hpcdata/software/containers/alphafold/alphafold_2.3.2.sh <OPTIONS>
    
    Required Parameters:
    -o <output_dir>         Path to a directory that will store the results.
    -f <fasta_file>         Path to a FASTA file containing one sequence
    ...

• Prepare Input: Set up necessary files for calculation

• Prepare Job Script: Create a job script with the commands to run the cluster. Here is an example of a alphafold.sh script

  #!/bin/bash
  #SBATCH --job-name=alphafold
  #SBATCH --output=alphafold_%j.out
  #SBATCH --error=alphafold_%j.err
  #SBATCH --nodes=1
  #SBATCH --cpus-per-task=16 # Adjust based on your system and needs
  #SBATCH --mem=64G       # Adjust memory as needed
  #SBATCH --time=24:00:00  # Adjust runtime as needed
  #SBATCH --gres=gpu:1     # Request a GPU
  
  module load alphafold # Or however you load the alphafold environment
  
  # Example command to run AlphaFold
  alphafold run_prediction \
    --fasta_paths=target.fasta \
    --output_dir=output_dir \
    --data_dir=/path/to/alphafold/data \
    --preset=model_1_ptm \
    --max_template_date=2023-12-31
  
  #Explanation of important parts:
  
  #SBATCH directives:
  #   --job-name: Name of the job.
  #   --output: Output file.
  #   --error: Error file.
  #   --nodes: Number of nodes.
  #   --cpus-per-task: Number of CPU cores per task.
  #   --mem: Memory allocation.
  #   --time: Maximum runtime.
  #   --gres=gpu: Number of GPUs requested.
  
  #module load alphafold: Loads the AlphaFold environment. This will vary depending on your HPC setup.

• Submit Jobs: Send your batch submission to start the calculation:

  user@login1:~$ sbatch alphafold.sh

• Monitor Progress: Check the status of your calculations

  user@login1:~$ squeue
               JOBID PARTITION     NAME     USER ST       TIME  NODES NODELIST(REASON)
               62847 partition   alphafold  user  R      13:05      1 vgpu-2017-001

  • JOBID: A unique numerical identifier for each job

  • PARTITION: The name of the partition (queue) where the job is submitted

  • NAME: The name assigned to the job

  • USER: The username who submitted the job

  • ST: The current status of the job: PD Pending, R Running, CD Completed, F Failed, S Suspended

  • TIME: The amount of time the job has been running

  • NODES: the number of noes allocated to the job

  • NODELIST(REASON): The names of nodes allocated to the job. If the job is pending, this column may display the reason why it’s waiting(e.g. “Resources”, “Priority”, “Dependency”)

• Analyze Results: Review results when they finish either on the HPC or back on your local computer for analysis and visualization.

1.3.1 Happy computing!