CSC - IT center for science

CSC services

research.csc.fi/en/service-catalog

Why use CSC services?

• CSC specialist support
• “Outsource” heavy/specialized computations
• Free of charge for open science Finnish universities and research institutes

Supercomputer

Main differences to own computer:

• Not faster, but bigger
• For speed up: parallelism
• Memory and CPU(/GPU) availability (application needs to make use of this!)
• Non-interactive for heavy computations
• Resource knowledge

Possibilities - supercomputer

• Use more memory/CPU/GPU than your own computer has available
  
  

  → analyse large files, Machine learning model training

  
  
• Speed up so called embarrassingly parallel analyses (many identical, but separate tasks)
  
  

  → doing same thing to multiple map tiles/ data chunks

Puhti supercomputer - Basics

Puhti supercomputer - Applications

Puhti supercomputer - r-env

• includes > 1100 R packages, including many geospatial packages
• Geospatial packages documentation
• General r-env documentation
• Running R scripts on Puhti- self learning examples

→ Instructions on additional R package installation in CSC docs

Puhti supercomputer - Data availability

Running your own R script in Puhti

0. Get CSC user account.
1. Log in to Puhti web interface (www.puhti.csc.fi).
2. Move your data and scripts to Puhti.
3. Open RStudio.
4. Check R package availability.
5. Fix paths of your input/output files.
6. Test your script with some test data.

…

Make use of the power of Puhti

…

7. Write a batch job script.
8. Run your scripts with all data as batch job (or interactively)
9. Make use of several cores using future package in your R code, if needed.

“My R code runs slow, what can be done?”

• Try to understand which part of the code takes time and why
  • Use system.time() or tictoc package
• Different R packages may provide same functions but are implemented differently (i.e. run faster/slower)
  • e.g. prefer sf over sp and terra over raster.
• Always be suspicious of for-loops!
• Consider parallelization
• Understand that number of cores != multiplier of speedup

Parallelization locally and on the supercomputer

Within R:

• use package future
• or snow, foreach, Rmpi,…

Outside R:

• GNUparallel
• other workflow tools can also be use, see CSC High Throughput Computing documentation

Training

• ‘CSC computing environment’ self-study course
• ‘Research data management’ self-study course
• CodeRefinery workshop September 2023
• STAC workshop
• Geocomputing with Puhti in Fall 2023
• CSC geoinformatics training materials

→ follow our training calendar

4. Support

docs.csc.fi

research.csc.fi

+ servicedesk@csc.fi

+ User support session in Zoom every Wednesday at 14.00

4.1 How we can help

Summary - Why use a supercomputer?

⌛ Resource needs (time, memory, storage, GPU)

👾 “Outsource” heavy computations, keep own computer free

🏘 Prebuilt environments, application availability

📊 Run many experiments at same time

🌐 Data availability

👥 Collaboration possibility

❓ CSC specialist support

💸 Free of charge for open science at Finnish universities and research institutes.

Getting started

Visit our Geocomputing page

• Step by step instructions
  1. Create personal user account
  2. Find a senior researcher / PI to create (a personal account for themselves and) and CSC project
  3. Ask above person to add you to the project and add needed services
• Find your account and project information
• Read the docs
• Check our tutorials and geocomputing examples