Overview of DeiC Interactive HPC at Aarhus University
SaaS • PaaS • IaaS
Department of Agroecology, AU Foulum, Wednesday, September 17th 2025
Your guides today
Agenda for today
- Organization
- Platform
- Access and Resources
- Help and Support
- Live demo
- some notes
The DeiC Interactive HPC consortium
Center for Humanities Computing
—
Faculty of Arts
The Interactive HPC Platform
Software
Hardware
Data security
Secure platform
-
UCloud is built from the ground up following rigorous security and privacy-by-design principles. All the systems at the eScience Center are tested against comprehensive security benchmarks by the Center for Internet Security (CIS).
-
The SDU eScience is ISO/IEC27001:2013 certified since February 2020.
-
AU has a framework agreement with SDU covering data processing of restricted data.
Development
Platform is Linux – mostly applications with either command line or web browser user interfaces.
- Development of the platform is done within the consortium
- Introducing new apps
Access and Resources
Interactive HPC is a service provider in the WAYF federation
Apply for resources
- Local sandbox (at the moment ~50K core-hours, ~[120; 3,200] GPU-hours, and 150GB)
- Resources from the local AU pool of resources – contact deicfrontoffice@au.dk
- Resources from the national sandbox via DeiC – contact deicfrontoffice@au.dk
- Resources from the twice yearly national calls for applications via DeiC – contact deicfrontoffice@au.dk
Let's have a look: Apply for Resources
Help and Support
Webpages
Support offices
Documentation
Live demo
\(\ldots\) some notes
- ERDA
- Parallel computing -- when a bigger machine makes no difference
Using ERDA
As described at SFTP access to your ERDA account. This is a quick example to show the possibility. In a real environment, I would use ssh-keys instead of a password, which is controlled in ERDA, UCloud, and the rclone setup.
Configure ERDA end point for rclone
🕰️ 14:13 ~
>: rclone config create erda sftp host io.erda.au.dk port 2222 user per@cas.au.dk pass "*** ENCRYPTED ***"
[erda]
type = sftp
pass = *** ENCRYPTED ***
host = io.erda.au.dk
port = 2222
user = per@cas.au.dk
🕰️ 14:14 ~
>: rclone lsl --max-depth 1 erda:
10244 2025-05-08 21:03:55.000000000 .DS_Store
30 2023-02-08 09:23:17.000000000 .default.css
238 2024-11-25 11:43:11.000000000 welcome.txt
Run application on UCloud with ssh server: Terminal
Configure UCloud endpoint in rclone
🕰️ 14:26 ~
>: tail -9 .config/rclone/rclone.conf
[ucloud]
type = sftp
host = ssh.cloud.sdu.dk
user = ucloud
port = 2496
Copy files
🕰️ 14:26 ~
>: rclone copy erda:welcome.txt ucloud:
🕰️ 14:28 ~
>: rclone lsl --max-depth 1 ucloud:
...
4623 2025-09-16 14:22:04.000000000 .zshrc
238 2024-11-25 11:43:11.000000000 welcome.txt
Parallel Computation
Single threaded vs multithreaded
- by hand
- chose a platform with native support, e.g., Wolfram Language, Julia, …
- Look for special methodologies or libraries in your preferred system
Single Machine vs Cluster
Your own cluster at UCloud
"Slurm"
Notice, that the slurm cluster connect nodes as part of the K8s cluster that are NOT connected via infiniband.
If you want to do a simulation of physics where you need to discretize space-time, then you would benefit more using infiniband to for much faster connections between contiguous nodes. For data science jobs, you don't need infiniband
Apache Spark
Other people's cluster
- GenomeDK
- LUMI, Leonardo, Gefion, Jupiter...
- •••
Thank you!
🤔