Why run R on an HPC cluster?

• HPC = high performance computing
• more resources: cores, memory, long runs
• one core not much faster than on a normal computer
  → parallelization to use many cores
• pre-installed software
  • R environment and packages

Overview of CSC’s computing services

Coming in 2026: Roihu

• replaces both Puhti and Mahti
• available in March 2026

More information: https://docs.csc.fi/computing/systems-roihu/

What happens to Puhti and Mahti?

• Puhti compute closes 1 month afterwards
• Puhti storage closes in July 2026
• Mahti closes in August 2026

Short introduction to supercomputers

Images from https://csc-training.github.io/csc-env-eff/

High performance computing cluster

A closer look at Puhti

SLURM job scheduler

R environment on Puhti & Mahti

• module r-env

module load r-env

• loads the latest R version available on Puhti:
  https://docs.csc.fi/apps/r-env/#available
• currently: R v. 4.5.1
• loading a specific R version:

module load r-env/451

r-env is a container-based module

• self-contained environment
  • limitations with using other modules on Puhti
  • combining R and Python
• RStudio Terminal panel: inside the container

R packages in r-env

• over 1600 packages installed
• packages of each R version date-locked to a specific date
  • avoid conflicts between versions
  • increase reproducibility
    • package versions only updated in a new R version
• avoid updating packages when installing new ones

Adding new packages

• default package directory is write-protected

1. install yourself for your project in projappl
   see: https://docs.csc.fi/apps/r-env/#r-package-installations
2. ask for a general installation for all users (email servicedesk@csc.fi)

Interactive R on Puhti

• RStudio: Puhti web interface (or ssh tunnelling)
• console R
  • compute node shell
  • sinteractive on terminal

module load r-env
start-r

Interactive R on Puhti

• get started, develop and test R scripts
• light or medium heavy interactive work up to a few hours
• uses fast local storage (NVMe) for storing temporary files (local disk)
• limitations on resources
  • RStudio struggles → move to batch jobs

Non-interactive R on Puhti: batch jobs

• R script (.R)
  • all R commands to be run
• batch job script (.sh)
  • reserves resources, loads modules, sets up environment
  • bash script with a specific format

Basic template for R batch job script on Puhti

Submitting batch jobs

• submitted on the login node
  • login node shell in the Puhti web interface
  • ssh on a terminal
• by default, output and error files go to the same folder where job was submitted

sbatch my_batch_job.sh

Status of a batch job

To view the status of the job:

squeue -u $USER
# or
squeue --me

To cancel a submitted job:
- job id is shown on the terminal when you submit the job

scancel <job_id>

Resource use: seff

When the job has finished, check the resources it used:

seff <job_id>