Skip to content

Best Practices for Working with HPC Systems at ICM UW

Archiving and Compressing Data

  • Regularly archive and compress files to free up disk space and reduce the number of files (important for file quota and disk quota limits).
  • ⚠️ ICM UW does not create backups of user data – you are responsible for maintaining your own backups.
  • Useful tools: tar, gzip, bzip2.
  • Compressing binary data is usually pointless (time-consuming and inefficient).
  • A tar archive (without the -z option) is the fastest solution for handling a large number of small files.

📖 Instructions:
- Storage quota limits
- File compression – tar and gzip


Backup and Synchronization

  • Regularly back up your data locally or in the cloud.
  • Tools:
  • rsync – for directory synchronization,
  • rclone – for working with Google Drive and other cloud services.

📖 Instructions:
- Archiving and compression
- Rclone documentation


Data Transfer

  • Files can be uploaded to and downloaded from the HPC systems via Open OnDemand.
  • Keep in mind that transfer speed is mostly limited by your local disk performance.

📖 Instruction:
- Open OnDemand – file transfer


Scalability Tests

  • Run scalability tests before starting large production runs.
  • Check if your program efficiently uses multiple cores and multiple nodes.
  • This will help you avoid wasting both resources and queue time.

📖 Instruction:
- SLURM basics


Secure Login

  • Try to configure SSH keys instead of passwords – this is both safer and faster.
  • Configure MasterConnection (SSH multiplexing) to avoid entering OTP repeatedly.

📖 Instructions:
- Creating and using SSH keys
- SSH multiplexing – overview


Resource Usage

  • Remember that large computations affect other users.
  • Do not attempt to bypass queue system limits – such actions may result in account suspension.
  • Unnecessary overuse of allocated resources decreases the efficiency of the entire system.
  • Although ICM does not charge for unused computing hours, you should carefully plan your requests for resources.
  • If your project requirements change, you may request additional resources within the same year.

📖 Instruction:
- Resource limits


Walltime and Job Queue

  • Accurately estimate the required walltime – it directly affects queue wait time and optimal use of resources.
  • The job scheduling system allows:
  • setting the number of jobs,
  • defining dependencies between jobs,
  • controlling the execution sequence.
    📖 Instruction:
  • SLURM Basics

Ta strona używa plików cookies.
Polityka Prywatności    AKCEPTUJĘ