Getting Started on the UL HPC platform
Copyright (c) 2013-2021 UL HPC Team <email@example.com>
This tutorial will guide you through your first steps on the UL HPC platform.
- make sure you have an account (if not, follow this procedure), and an SSH client.
- take a look at the quickstart guide
- ensure you operate from a Linux / Mac environment. Most commands below assumes running in a Terminal in this context. If you're running Windows, you can use MobaXterm, Putty tools etc. as described on this page yet it's probably better that you familiarize "natively" with Linux-based environment by having a Linux Virtual Machine (consider for that VirtualBox) or Windows Subsystem for Linux.
From a general perspective, the Support page describes how to get help during your UL HPC usage.
In the below tutorial, you'll proposed terminal commands where the prompt is denoted by
In general, we will prefix to precise the execution context (i.e. your laptop, a cluster frontend or a node). Remember that
# character is a comment. Example:
# This is a comment $> hostname (laptop)$> hostname # executed from your personal laptop / workstation (access-iris)$> hostname # executed from access server of the Iris cluster
You can find a brief overview of the platform with key characterization numbers on this page.
The general organization of each cluster is depicted below:
Details on this organization can be found here
Discovering, visualizing and reserving UL HPC resources
In the following sections, replace
<login> in the proposed commands with you login on the platform (ex:
Step 1: the working environment
After a successful login onto one of the access node (see Cluster Access), you end into your personal homedir
$HOME which is shared over GPFS between the access node and the computing nodes.
Otherwise, you have to be aware of at least two directories:
$HOME: your home directory under NFS.
$SCRATCH: a non-backed up area put if possible under Lustre for fast I/O operations
Your homedir is under a regular backup policy. Therefore you are asked to pay attention to your disk usage and the number of files you store there.
Estimate file space usage and summarize disk usage of each FILE, recursively for directories using the
You can get an overview of the quotas and your current disk usage with the following command:
You shall also pay attention to the number of files in your home directory. You can count them as follows:
(access)$> df-ulhpc -i
Step 2: web monitoring interfaces
Each cluster offers a set of web services to monitor the platform usage:
Ganglia is a scalable distributed monitoring system for high-performance computing systems such as clusters and Grids. Ganglia provides plots the system usage for each individual compute nodes (CPU, memory, I/O and network usage).
These information will help you identify and understand the behavior of your jobs on the cluster.
It is interesting to identify the limiting factor of your job:
- Storage I/O
- Network bound
This is covered in the other tutorial Monitoring and profiling
SLURM Web is a website that show the status of jobs and nodes with a nice graphical interface.
You can use the SLURM web interface for a visualization of the job scheduler state (list and gantt chart visualization).
Sample Usage on the UL HPC platform: Kernel compilation
We will illustrate the usage of GNU screen by performing a compilation of a recent linux kernel.
start a new screen session
rename the screen window "Frontend" (using
create a new window and rename it "Compile"
within this new window, start a new interactive job over 1 node and 2 cores for 4 hours
(access)$> si --time 2:00:0 -N 1 -c 2
detach from this screen (using
- kill your current SSH connection and your terminal
- re-open your terminal and connect back to the cluster frontend
list your running screens:
(access)$> screen -ls There is a screen on: 9143.pts-0.access (05/04/2014 11:29:43 PM) (Detached) 1 Socket in /var/run/screen/S-svarrette.
re-attach your previous screen session
(access)$> screen -r # OR screen -r 9143.pts-0.access (see above socket name)
in the "Compile" windows, go to the temporary directory and download the Linux kernel sources
(node)$> cd /tmp/ (node)$> curl -O https://cdn.kernel.org/pub/linux/kernel/v4.x/linux-4.19.163.tar.xz
IMPORTANT to avoid overloading the shared file system with the many small files involves in the kernel compilation (i.e. NFS and/or Lustre), we will perform the compilation in the local file system, i.e. either in
/tmp or (probably more efficient) in
/dev/shm (i.e in the RAM):
(node)$> mkdir /dev/shm/PS1 (node)$> cd /dev/shm/PS1 (node)$> tar xf /tmp/linux-4.19.163.tar.xz (node)$> cd linux-4.19.163 (node)$> make mrproper (node)$> make alldefconfig (node)$> make 2>&1 | tee /dev/shm/PS1/kernel_compile.log
- You can now detach from the screen and take a coffee
The last compilation command make use of
tee, a nice tool which read from standard input and write to standard output and files. This permits to save in a log file the message written in the standard output.
Question: why using the
make 2>&1 sequence in the last command?
Question: why working in
/dev/shm is more efficient?
- Reattach from time to time to your screen to see the status of the compilation
Your compilation is successful if it ends with the sequence:
[...] Kernel: arch/x86/boot/bzImage is ready (#2)
Restart the compilation, this time using multiple cores and parallel jobs within the Makefile invocation (
-joption of make)
(node)$> make clean (node)$> time make -j $SLURM_CPUS_ON_NODE 2>&1 | tee /dev/shm/PS1/kernel_compile.2.log
The table below should convince you to always run
make with the
-j option whenever you can...
- Use the Ganglia interface to monitor the impact of the compilation process on the node your job is running on.
Connect to your interactive job using the command
sjoin <jobid>. Use the following system commands on the node during the compilation: