Introduction to LSU HPC

High-throughput data needs high-throughput analysis

• Large volumes of data
• Large number of chains or iterations for model simulations (theoretical or statistical)
• Large number of repetitions

Image from https://ourworldindata.org/grapher/supercomputer-power-flops?time=earliest..2024

HPCs can be hard, we are scratching the surface

At LSU, we have access to several HPCs

Allocations on the HPC

• To use the HPC, each user account must have access to at least one computing allocation
  • All of you are on the course allocation, but beyond the semester, ask your PI to create an allocation
  • Each PI by default can have two active “startup” allocations (150,000 CPU hours), and can always submit a proposal for more.

How to log into the HPC?

• Through an SSH connection in your terminal or through a new web-based interface

• Connecting through the terminal is ideal for large, production-scale work, but the web interface is fantastic for a lot of prototyping (and maybe more, depending on your needs)

• Terminal: through your terminal (mac/linux) or GitBash (Windows), you can access your LSU HPC account with: ssh <username>@mike.hpc.lsu.edu. (Needs 2FA setup).

• Web-based interface: https://ondemand.mike.hpc.lsu.edu

What happens when you log into the HPC?

What happens when you log into the HPC?

Once you are logged in, you need to request access to a “compute node”

• How much time do you want on the node?
• What allocation should these hours be charged to?
• How many nodes and how many cores do you need? (Depends on what kind of work you want to do!)
• You can request an interactive node or a job node – pros and cons to each

Requesting compute access

• Web access: simple on the portal

What happens when you log into a compute node?

File organization on the HPC

File organization on the HPC

File organization on the HPC

• You can connect the HPC to git just as you did with your laptops
  • ssh-keygen on the HPC terminal; copy pubkey into gitlab
  • Useful if you want to prototype code on your laptop on a subset of the data and then ship it off to the HPC, or vice-versa

What if you want to do more?

• Accessing the HPC through the terminal opens up more options

• Upon logging in, you will be on the head node

• We need to request access to a compute node to do any work:

What if you want to do more?

• Once on a compute node, you can load specific modules to achieve tasks

• e.g. Bowtie, SamTools, RevBayes

• Once a module is loaded, you can use it directly from the command line as you would on your on terminal.

What if you want to do more?

• So far, we have covered Interactive sessions for interacting with the HPC
  • This means you are sitting at the keyboard, telling the computer what to do after it completes each step
  • Some limitations to this!
• Instead, we can submit job batches

Batch job submissions

• LSU HPC uses a job management program called SLURM (“Simple Linux Utility for Resource Management”) to manage its tasks

• SLURM intelligently decides how to divide up available computing power among jobs, in the order they were submitted (or according to other priorities)

• Users can submit a bash script that contains a list of tasks:

 #!/bin/bash
 #SBATCH -N 1               # request one node
 #SBATCH -t 2:00:00         # request two hours
 #SBATCH -p single          # in single partition (queue)
 #SBATCH -A your_allocation_name
 #SBATCH -o slurm-%j.out-%N # optional, name of the stdout, using the job number (%j) and the hostname of the node (%N)
 #SBATCH -e slurm-%j.err-%N # optional, name of the stderr, using job and hostname values
 # below are job commands
 date

 # Set some handy environment variables.

 export HOME_DIR=/home/$USER/myjob
 export WORK_DIR=/work/$USER/myjob
 
 # Make sure the WORK_DIR exists:
 mkdir -p $WORK_DIR
 # Copy files, jump to WORK_DIR, and execute a program called "mydemo"
 cp $HOME_DIR/mydemo $WORK_DIR
 cd $WORK_DIR
 ./mydemo
 # Mark the time it finishes.
 date
 # exit the job
 exit 0

Why use jobs?

• You need not be at the computer – the program will work through its activities one by one

General workflow for HPC

• Decide if your work needs the use of an HPC
• Decide if you want to work interactively or through job submissions
• Decide if the web-based portal is enough for your work, or if you need to turn to the terminal
• Always test code on subsets of the data/iteratively
• Think about “minimum working examples” or “minimum reproducible examples”