HoloBaseEvolutionary Hologenomics Training Hub Open the course

Draft for review — preview version 4b86b29. HoloBase teaching material, not yet open to students and not part of the Mjolnir documentation.

Review notes · draft

Module 00 — Mjolnir Foundations

Thank you for reviewing this draft. It is a shared HoloBase course that every student completes once before their first real Mjolnir session. Below are the Mjolnir-specific statements it makes, where each comes from in the Mjolnir Knowledge Base, how the practice simulator simplifies things, and the questions we'd most like your view on.

What this is — and isn't

  • HoloBase teaching material based on your Mjolnir Knowledge Base — not official Mjolnir documentation. It credits you and the Knowledge Base, and links the page each exercise draws on.
  • A course of 25 short exercises with a browser-only simulation of a Mjolnir session: a fictional user (sim123), project (hb_training) and compute nodes (simnode01–simnode04).
  • The simulation never connects to Mjolnir or anywhere else. The page's security policy blocks every network connection from its scripts, and it runs no real commands.
  • It does not reproduce First 15 minutes on Mjolnir. Students are sent there for their real first session, after their account is approved through Request access.
  • Every exercise links the Knowledge Base page it is based on, and the course says the Knowledge Base takes precedence if anything disagrees.

Questions for you

  1. Accuracy. Is any statement in the table below out of date, or worded in a way you'd change?
  2. Paraphrases. Four Knowledge Base rules are paraphrased rather than quoted (marked ≈ in the table): login-node use, the home folder, project storage and fairshare. Does each paraphrase keep your meaning?
  3. Gaps. Is there anything every new user should understand before their first session that the course leaves out — for example SSH keys, or the shared databases in /datasets/globe_databases?
  4. The practice prompt. The simulator shows the real login-node name mjolnirgate, clearly labelled SIMULATION. Is that acceptable, or would you prefer a fictional name?
  5. Simplifications. Are the simplifications listed below acceptable for teaching, or would any of them mislead a new user?
  6. Next phase. A short readiness check (about eight questions) will follow. Would you like to review those questions too?

Please send your comments to the HoloBase team member who shared this preview with you — by reply, or as notes against the exercise numbers.

Mjolnir-specific statements and their sources

Checked against the Knowledge Base on 2026-10-22. ≈ marks a rule the course paraphrases rather than quotes.

Statement in the course or simulatorKnowledge Base page
Login node mjolnirgate; connect with ssh <KU-ID>@mjolnirgate.unicph.domain over the UCPH VPNFirst 15 minutes; VPN and SSH access
≈ Computation isn't allowed on the login node; repeated misuse can mean losing accessLogin nodes vs compute nodes
--qos is mandatory; jobs without a QoS are rejectedPartitions, QoS, and fairshare; Submitting jobs
Partitions cpuqueue (default), gpuqueue, lazyqueue, plus filetransfer and teaching (restricted); defaults 1 hour and 2 GB per CPU; QoS normal max 48 CPUs; lazyqueue with --qos=lazyPartitions, QoS, and fairshare
≈ Fairshare counts what you request, not what your job uses; small, realistic requests backfill more easilyPartitions, QoS, and fairshare
Reply "Submitted batch job <jobid>"; squeue -u $USER, states PD / R / CG; squeue -u $USER -o "%.10i %.8T %R"First 15 minutes; Monitoring jobs
sacct -j, scontrol show job, scancel, seff (layout of the illustrative example)Monitoring jobs; Check job efficiency; Cancelling jobs
srun --partition=cpuqueue --qos=normal --nodes=1 --cpus-per-task=2 --mem-per-cpu=2G --time=01:00:00 --pty bash; idle sessions still hold resourcesInteractive jobs
Environment Modules 4.5.2, not Lmod; module spider doesn't exist; module avail/load/list/unload/switch/purge/display/whatis/apropos; samtools/1.21Find and load software
≈ /home/<KU-ID> limited to 100 GB and not meant for datasets or job output; /projects/<project>/people/<KU-ID>/, data/, apps/, scratch/ (not backed up)Where should my files go?
Don't create a missing project or people folder — ask through Getting helpFirst 15 minutes
du -sh --apparent-size for reporting; du -h --apparent-size --max-depth=1 … | sort -hHow much storage am I using?
≈ Project storage is shared, finite and paid for; no enforced quota doesn't mean unlimited storageStorage limits and responsibilities
Support Center, and what to include in a request; new accounts and VPN problems go elsewhereGetting help
Accounts: User application form, confirmed by the PIRequest access

How the simulator simplifies things

  • Time runs faster in the simulation (about five times faster while a job runs), and jobs pend for a few seconds before starting.
  • Error and status messages are the simulator's own explanations, labelled "Simulation:". Where the Knowledge Base gives no exact text (for example the rejection for a missing QoS), none is invented.
  • Default output file slurm-<jobid>.out when --output is omitted is general Slurm behaviour; the Knowledge Base doesn't state it.
  • Module catalogue: only samtools/1.21 comes from the Knowledge Base. Other names and versions (samtools/1.17, fastqc, multiqc, bwa, python) are examples, and the page says so.
  • Pending reasons (Priority) and (Resources), sacct values, seff figures and file sizes are illustrative.
  • Compute node names are fictional, because the Knowledge Base doesn't name compute nodes.
  • Not simulated: zcat/gzip, salloc (pointed to srun --pty bash), sinfo, GPUs, tail -f (shows the file once), less (prints the whole file), vim (opens the simple editor).
  • Safety stops: rm on the home, people or shared folders is stopped with an explanation. Real computation on the login node is stopped with the Knowledge Base rule.

Open the course