Exercise 0 - Reinforce what was seen in the
introduction
Authors: BIOI2 & I2BC members
Last updated: 2025-03-13
Objectives
In this Exercise, we will be taking our first steps with the cluster.
We will:
connect to the cluster through ssh
understand the cluster architecture and resources available on the
cluster (slave nodes, CPUs, memory, etc.)
run your first (interactive) job
learn how to access the files from “partages”
get familiar with the computing spaces
learn how to find the software you need for your analysis
⁕ ⁕ ⁕ ⁕ ⁕ ⁕
Setup
Task 1: Connect to the cluster
Connecting to the cluster through the command line is a two-step
process. You should first connect to a “bridge server” called
passerelle, then to the actual cluster from the
passerelle using the ssh (=“Secure SHell”)
command.
The addresses are:
passerelle: passerelle.i2bc.paris-saclay.fr
cluster: slurmlogin.calcul.i2bc.paris-saclay.fr
Click to see hints
you need a terminal
Windows: open a Windows Power Shell (open the Start menu and type
windows powershell in the search bar)
Linux: open a terminal (open the Activities menu and type terminal
in the search bar)
Mac: open a terminal (in the Finder, open the
/Applications/Utilities folder, then double-click Terminal)
NB: you can also use third party software such as MobaXterm, X2Go or
TurboVNC…
the ssh command is used as follows
ssh login@passerelle.i2bc.paris-saclay.fr
Login & password are your MultiPass credentials
(usually firstname.lastname). NB: Don’t be surprised if the characters
of your password don’t appear on the screen while you’re typing them,
that’s normal. Just press Enter to validate.
Click to see answers
Connect to the passerelle
The passerelle is a “bridge server” that you can use in order to
connect to most other servers of the I2BC, including the cluster. Use
the ssh command to connect to it:
ssh john.doe@passerelle.i2bc.paris-saclay.fr
Login & password: your MultiPass credentials
(usually firstname.lastname). Don’t be surprised if the characters of
your password don’t appear on the screen while you’re typing them,
that’s normal.
Connect to the cluster
Use the ssh command to connect to the cluster from the passerelle
(you can also bypass the passerelle if you’re directly connected to the
I2BC network):
ssh slurmlogin.calcul.i2bc.paris-saclay.fr
Password: same as before.
You’re now on the Master node of the cluster. It’s called
“slurmlogin” at the I2BC and it uses Debian with a bash shell.
Tip: Where am I?
Each line in the terminal starts with a prompt, for example:
john.doe@slurmlogin:/home/john.doe$
who you are:
John Doe
on what you’re
connected: slurmlogin (the master node)
where you are: in
John’s home directory
⁕ ⁕ ⁕ ⁕ ⁕ ⁕
Computational resources
Question 2: What are the 4 main components of the
I2BC computer cluster?
Click to see answer.
If we simplify, the I2BC cluster is made of the following:
master node or “login node” (slurmlogin)
slave nodes or compute nodes (e.g. node01)
scheduler which manages the resources (slurm)
storage/computing spaces
Question 3: Why should I avoid running programmes
on the master node?
Click to see answer.
/!\ the master node is not made for heavy computation. You should
always run your programmes on one of the slave nodes (=the workers)!
That’s because it’s used as a login node and has limited resources.
Thus, running programmes or scripts directly on it will slow down the
cluster for everyone.
Question 4: Are all slave nodes the same?
Click to see answer.
Yes and no :-)
All nodes are alike but the amount of resources might be different
(e.g. RAM memory, number of CPUs/GPUs…)
Each node belongs to one or several groups, called
partitions. The default partition is “common” and
regroups all the communal nodes (i.e. nodes that were bought by the I2BC
and that are accessible by all).
Other partitions exist to regroup nodes with specific features or
nodes belonging to a specific team.
Have a look at the partitions
tab on the intranet for more information on the different partitions
that exist on the I2BC cluster.
Task 5: list all the resources on the I2BC cluster
with sinfo
Running sinfo -s will list a more summarised version of
all resources.
Click to see the output of the above command.
PARTITION AVAIL TIMELIMIT NODES(A/I/O/T) NODELIST
common* up infinite 2/17/1/20 node[06-07,13-16,21-24,27-28,30-37]
alphafold up infinite 0/1/3/4 node[41,49-51]
smallgpu up infinite 0/1/0/1 node38
run2 up infinite 2/17/1/20 node[06-07,13-16,21-24,27-28,30-37]
run4 up infinite 2/17/1/20 node[06-07,13-16,21-24,27-28,30-37]
run8 up infinite 2/17/1/20 node[06-07,13-16,21-24,27-28,30-37]
run16 up infinite 2/17/1/20 node[06-07,13-16,21-24,27-28,30-37]
run32 up infinite 2/17/1/20 node[06-07,13-16,21-24,27-28,30-37]
lowprio up infinite 2/19/1/22 node[06-07,13-16,21-24,27-28,30-38,40]
lowpriogpu up infinite 0/2/0/2 node[38,49]
sinfo -s gives you a list of partitions and the nodes
that are part of each partition. As you can see, some nodes belong to
several partitions.
The NODES column indicates the number of nodes that are
available/idle/other/total within each partition, available+idle being
the total number of nodes that are functional (idle meaning the node is
completely free = no jobs running on it at the moment).
⁕ ⁕ ⁕ ⁕ ⁕ ⁕
Storage & computing spaces
When you start using the cluster, it’s important to distinguish
between storage and computing spaces and in what context you should use
one or the other.
On the cluster, you have access to your usual “partages” spaces
(=/store) from any of the nodes. And you also have access
to more cluster-specific spaces such as /data/work/I2BC,
your /home and the /scratchlocal specific to
each node.
Different spaces should be used for different things:
Path
Function
➜ /store/EQUIPES
- raw data &
final processed data (for safe keeping) - all data
& protocoles that should be accessible by the team
and that cannot be easily re-generated
➜ /store/USERS
- personal but
professional data (e.g. course material for PhD students or teachers,
work contracts etc.)
- temporary data that
needs to be accessible from all nodes (e.g. a copy of a small database
fetched from the internet, non-dividable working files, etc.) - data
that is copied from somewhere else (e.g. for sharing
with other teams)
➜ /scatchlocal
- temporary data that
doesn’t have to be accessible from all nodes (e.g. intermediate files
generated by a tool)
Question 6: I generated data for my team’s project,
where should I store it?
Click to see answer.
Anything linked to your team’s project is best stored within the team
space on “partages” i.e. in /store/EQUIPES/your_team/.
Why? so that everyone of your team has access to it, even
after you leave.
Of note, every team member also has a dedicated space within
/store/EQUIPES/your_team/MEMBERS/ which is readable by
everyone but writable only by him/her and the team’s PI.
Question 7: My script generates plenty of temporary
intermediate files, where should they go?
Click to see answer.
You have 2 options:
/data/work/I2BC/: everyone of the I2BC has read &
write access, it’s a space of 20 To total
/scratchlocal/: everyone of the I2BC has read &
write access for there own files & folders, it’s a space that is
specific to each node, sizes vary from one node to another
Both spaces aren’t cleaned automatically, so it’s up to you to do
it!
If you have a lot of read & write operations to do, it might be more
interesting to use the /scratchlocal/ space which is local
to each node (avoids saturating the network).
Question 8: Why shouldn’t I store temporary files
within the “partages” spaces (/store)?
Click to see answer.
/store (“Partages”) and /home are
backed-up whereas /data/work/I2BC and
/scratchlocal are not
backed-up = you can fetch old versions of your
files from up to 3 months ago
Thus the data you put on “Partages” spaces is stored for up to 3
months (even after deleting it). Saving temporary files to it will take
up a lot of space for nothing.
NB: backups are not specific to the cluster, you can
access them:
from within the cluster, through the hidden .snapshots
folder, from anywhere within the backed-up file system (you can use
ls to navigate through it and cp to copy old
files accross).
/!\ just remember, for temporary spaces, delete your data when
you’ve finished because it’s not done automatically and space is
limited!
-C is an option to filter outputs to only those
containing the given search term-i is an option
to make the search case insensitive
Task 10: Load FastQC and run
fastqc --version to see if it worked
/!\ Warning: Most softwares are not accessible on the master node
as it’s not made for running scripts and programmes, so make sure you
connect to a slave node first!
Click to see answer.
module load fastqc/fastqc_v0.11.5 won’t create an error
if you run it from the master node. However, if you try running
fastqc, it won’t work. Thus, we first have to connect to a
salve node.
If your software isn’t on the cluster, contact the
SICS, they’ll install it rapidly for you
⁕ ⁕ ⁕ ⁕ ⁕ ⁕
Interactive session
Task 11: Connect, then disconnect from a node of
the cluster.
If you want to use some resources, you have to ask the Scheduler
using a specific syntax: the Slurm
language (this might be different on different clusters but
Slurm is quite widely spread).
Hints:
To connect to a node in interactive mode:
srun --pty bash
