FAQ
What the CERN EOSC Node is, what it costs you (nothing), and why running your analysis here, next to the data and software, is different from starting from scratch on a commercial notebook service.
What it is and getting on
What is the CERN EOSC Node?
The CERN EOSC Node is a place to run real particle-physics analysis in your browser, on CERN infrastructure, next to the data and software you need. It is part of the EOSC Federation, the European effort to connect national and institutional research infrastructure into one federated system.
Who can use it?
Researchers can log in through federated login, so you do not need a CERN computing account. We are still finalising which identity providers the Node accepts. The Get Started page has the current login path.
How do I start?
Go to Get Started. It shows three paths: run a ready-made analysis with one click, bring your own notebook and work against Open Data, or package your analysis as a reproducible REANA workflow. All three use the same federated login.
Is it free, and what is the catch?
Yes, using the Node costs nothing. It is shared public infrastructure with finite capacity, not an unlimited commercial cloud. Fair-use limits apply. It is for research and teaching, not production hosting.
Resources and performance
What resources do I get in a session?
A working session is comparable to 2 CPU and 8 GB of RAM. We measured the timings on this page on that setup. We are still finalising the exact quota, and whether it differs between the CERN VRE and REANA.
Are there time or idle limits?
We have not set them yet. We will publish them here when we do. For comparison, Google Colab's free tier disconnects after about 90 minutes of inactivity and 12 hours in total. MyBinder ends a session after 10 minutes idle.
How many people can use it at once?
Capacity is shared. We are increasing it in phases toward the Node's production milestone in October. We have not published a concurrency figure yet. The aim is a fair share per user, not unlimited simultaneous load.
Can I get a GPU?
Not for now. The Node offers CPU and memory only. We may review this later, depending on demand. We measured every timing on this page, including the ones below, on CPU alone.
How long will my analysis take?
One 10,000-event Open Data file runs through Rivet in about 4 minutes on a single core, roughly 39 events per second. A typical 100,000-event study is about 20 minutes on two cores, around 0.7 CPU-hours in total. The scaling is roughly linear. To estimate your own job, count the events: ten times the events takes roughly ten times as long.
How does this compare to Google Colab or MyBinder?
Different tool for a different job. Colab gives you more memory (about 13 GB) and MyBinder gives you less (1 to 2 GB), but on both you start from an empty machine: you install the physics software yourself and download every data file across the network, every time you connect. MyBinder's disk and idle limits mean it cannot even hold a single 2.1 GB Open Data file for the length of one event loop.
On the Node the data is already mounted (EOS) and the software is already there (CVMFS), so the same analysis starts immediately and reads data locally instead of over the network.
| Platform | CPU | RAM | Idle / max session | Note |
|---|---|---|---|---|
| The Node (measured) | 2 CPU | 8 GB | to be published | EOS + CVMFS: data and software already there |
| Google Colab (free) | 2 vCPU | about 13 GB | about 90 min idle, 12 h max | no CVMFS, data downloaded over the network |
| MyBinder (free) | up to 1 CPU | 1 to 2 GB | 10 min idle cull | cannot hold a single 2.1 GB Open Data file |
Colab and MyBinder figures are the platforms' published free-tier defaults, which both vary without notice. We measured the Node figures on the session used for the timings in the previous question.
Data
What data can I use?
CERN Open Data: real and simulated collision data released for public use. It includes the ATLAS, CMS and LHCb datasets in the Analysis Catalogue and the 13.6 TeV event-generation release that we used for the timings above. Browse it directly, or launch straight into a ready-made analysis.
How fast is data access compared to downloading it myself?
This is the main practical difference from a commercial free tier. Open Data volumes are mounted directly on the Node (EOS). You read a 2.1 GB file in place, with no download and no egress fees. On a commercial cloud you pay for that transfer in time, and sometimes in money, on every single run.
Can I bring my own data or code?
Yes: bring your own notebook to the CERN VRE and work against Open Data with your own code. Rucio manages larger private datasets across sites. We are still working out the upload limits and the exact policy for external users' private data.
Do I keep my files between sessions?
Interactive work in the CERN VRE and REANA is backed by CERNBox, so your notebooks and files persist between sessions. Do not use it for long-term storage. We are still finalising how long we keep scratch.
Software and reproducibility
What software is already installed?
The CERN and LHC analysis stack, delivered through CVMFS: ROOT, Rivet, atlasopenmagic, and the usual Python scientific stack. On Colab or MyBinder none of this is preinstalled and CVMFS is not available. You install and build the stack yourself, every time.
Can I install my own packages?
Yes, pip install works normally inside your session. It installs into your own user environment, not the shared service. Other users and the base stack are not affected.
Will my analysis be reproducible?
Yes, and this is a real advantage over an ad-hoc notebook on a commercial free tier: package your analysis as a containerised REANA workflow. Anyone can re-run it and get the same result, because the environment is captured with the code.
Trust, scope, support
Is my session isolated from other users, and is it secure?
Interactive sessions run in a network segment isolated from the rest of CERN's infrastructure. An incident in one session cannot spread to other services or users. Access requires authentication through your federated login. There is no anonymous access to compute.
What is the Node good for, and what is it not for?
Good for: interactive analysis of Open Data, prototyping, teaching, and reproducible workflows built on the tools above. Not for: unlimited compute, long-running production services, or a general-purpose replacement for a commercial cloud account. The Node is a production-minimum research node, not a hyperscaler.
Can I use it to teach a course or run a workshop?
Yes: training and workshop use is a supported use case, planned for at the scale of a few hundred participants. Contact the Node ahead of your session so the team can prepare.
Where do I get help?
The Contact page is the way to reach the team. It lists the support channels and says which one suits your request. Get Started has guided paths if you are not sure where to begin.
What is on the roadmap?
We are expanding capacity and features in phases toward the Node's production milestone in October. We will publish details here when they are confirmed.
Still stuck?
If this page does not answer your question, the Get Started page has guided paths for researchers, policy makers, and infrastructure operators, and the Contact page can point you in the right direction for anything not covered here.
