Euler CPU Track

Complete this track before regular CPU computation on Euler. It is also a prerequisite for the Euler GPU track.

Prerequisites#

• IT and Research Safety Core
• Supervisor confirmation that Euler is appropriate for the project
• Working ETH account and VPN access when outside the ETH network

Competencies#

You can:

• distinguish login nodes, compute nodes, and storage systems;
• connect using a personal account and safe SSH configuration;
• submit, inspect, and cancel jobs with sbatch, squeue, myjobs, and scancel;
• inspect completed jobs with sacct and seff;
• use sstat appropriately for a running job step;
• interpret states, pending reasons, logs, and exit codes;
• start with conservative CPU, memory, and wall-time requests;
• optimize later requests from measured usage;
• preserve and record the source, configuration, and input versions used by a queued or running job;
• submit arrays with an explicit concurrency cap and calculate aggregate concurrency across all active submissions;
• distinguish the public and es_fuge accounts;
• choose $HOME, /cluster/work/fuge, $SCRATCH, and $TMPDIR intentionally;
• prefer sbatch for unattended work and short allocations for debugging.

Required Learning#

1. Euler access and SSH
2. Euler storage
3. Slurm job lifecycle
4. First Euler job
5. Resource optimization
6. Safe job arrays

Evidence#

• A tiny successful CPU job using one CPU and modest memory.
• Sanitized sacct and seff output for its job ID.
• An explanation comparing requested and measured resources.
• A correct explanation of the batch script snapshot versus external files.
• A corrected Slurm array with a justified concurrency cap.
• Correct storage choices for code, data, logs, checkpoints, and temporary files.

Understand Before Accepting AI Output#

• I inspected the resource request before submission.
• I know what will execute and where output will be written.
• I can cancel the job if the request or behavior is wrong.
• I measured usage before increasing resources.

Pass Criteria#

The student completes the practical evidence and correctly explains the job lifecycle. Memorizing Slurm flags without interpreting accounting output is not sufficient.