Laptop, NAS, Blade, And Euler | IDEAL IT Passport

Goal: choose the correct system before installing software or copying data.

Prerequisites: ask your supervisor where the project's source data and final deliverables belong.

Execution context: reading and decision exercise. No commands are required.

The lab uses several computers and storage services because no single system is suited to every task. A service is something you connect to, such as GitHub or the NAS. A computer is where a program actually runs, such as your laptop, Blade, or an Euler compute node.

The terms used below mean:

a GUI (graphical user interface) is an application controlled mainly with windows, menus, and buttons;
a CPU is the general-purpose processor used by most programs;
a GPU is an accelerator that helps only software written to use it;
a cluster is a group of shared computers managed as one service;
a job is a program plus its requested CPU, memory, and time, submitted to a scheduler; Slurm is Euler's scheduler;
durable storage is intended to keep required files under an approved ownership and recovery plan; temporary storage may be purged;
high input/output (high-I/O) work repeatedly reads or writes a large amount of data.

Decision Table#

System	What it is	Use it for	Do not use it for
Your laptop	The Windows, macOS, or Linux computer in front of you	Editing, small tests, meetings, local development	Sole copy of important research data
GitHub	A website that stores Git project folders, their file history, and reviews	Source code, configuration, tests, small documentation	Secrets, large datasets, checkpoints, generated results
IDEAL Lab NAS	Network-attached storage: the lab's durable shared project drive	Shared project data and durable team deliverables	Running high-I/O Euler jobs directly against NAS
Blade Windows server	A shared Windows computer reached through Remote Desktop	Graphical access to licensed Windows engineering software and short prototypes	Primary long-running machine-learning compute or permanent files on `C:`/`D:`
Euler	ETH Zurich's shared high-performance computing cluster, managed by the Slurm scheduler	Long, parallel, memory-intensive, or GPU computation submitted as jobs	Heavy computation directly on a login node, the entry computer used for files and job control
Euler `$HOME`	Your small private home folder on Euler; `$HOME` is a shell name for that path	Code, configuration, and small environments	Large datasets, checkpoints, or high-I/O workloads
Euler `/cluster/work/fuge`	The lab's large shared working storage on Euler	Large working data and artifacts	Files that should be source-controlled in personal clones
Euler `$SCRATCH`	Your temporary high-throughput Euler storage; `$SCRATCH` is a shell name for its path	Temporary large data that can be recreated	Only copy of important data; files are purged
Euler `$TMPDIR`	A shell name for temporary storage on the compute node running one job	Fast intermediate files needed only during that job	Anything needed after the job exits unless copied back

Compute Decision#

Use your laptop for editing and small smoke tests.
Use Blade when you need its licensed Windows GUI software.
Use Euler CPU nodes for long, parallel, memory-intensive, or unattended computation.
Use Euler GPU nodes only when the program actually executes GPU operations.
Use sbatch, Slurm's job-submission command, for unattended Euler work. Use euler-tunnel only for short interactive development after Slurm has assigned a compute node.
For ParaView, run the matching 6.0.0 GUI locally and connect it to a CPU pvserver allocation by following the ParaView client/server procedure.

Verification#

For an optional private project discussion, complete the location table below with its information owner. The current Passport mission checks fictional decisions; it does not approve or collect this real map.

Before starting a project, complete this table with your supervisor:

Source code:
Source data:
Working data:
Temporary data:
Checkpoints:
Final results:
Long-term archive:

Common Failure And Safe Recovery#

A large dataset is already in Git or $HOME. Stop adding copies. Ask the repository/data owner where it belongs, move it using an appropriate transfer method, verify the destination, then remove the unwanted copy safely. If it entered Git history, coordinate cleanup rather than rewriting shared history alone.

Ask For Help When#

the project has no declared storage owner;
data is confidential or contract-restricted;
you need to transfer terabytes;
collaborators cannot read or write the approved shared folder;
the only available copy is in temporary storage.

Continue with Incidents And Getting Help.