AI and agents2 min read

CERN is moving 2,200 machines from RHEL to Debian 13

RHEL 10 needs x86-64-v3, which would rule out nearly two thirds of CERN's accelerator controllers. So 2,200 industrial machines are moving to Debian 13.

CERN is moving 2,200 machines from RHEL to Debian 13

CERN, the organisation that runs the Large Hadron Collider, was for decades an almost textbook Red Hat shop. Together with Fermilab it built Scientific Linux, and after that CERN ran on CentOS and RHEL-like infrastructure.

But now CERN is moving more than 2,200 industrial computers and embedded systems, the ones that control the accelerators and the equipment connected to them, to Debian 13.

And this is not an “admins got tired of yum and fancied apt” story.

The final blow came from Red Hat’s policy on minimum CPU requirements:

  • RHEL 9 requires x86-64-v2;
  • RHEL 10 now requires x86-64-v3.

For an ordinary office: fine, just replace the machines. For CERN it means replacing a large number of specialised controllers that work next to the accelerators, depend on unique boards and have to run reliably for fifteen years or so.

RHEL 9 already ruled out about 47% of CERN’s existing front-end machines. Moving to RHEL 10 would have cut off roughly another 17%. Together that is nearly two thirds of the fleet (about 64%), which would have had to be replaced not because it could no longer do the job, but because the distribution no longer wants to run on its CPUs.

CERN called this forced obsolescence. And, frankly, it is hard to disagree: if a controller reliably runs complex industrial equipment, redesigning it for the sake of a new baseline ISA is an expensive and rather strange kind of progress.

Debian turned out to be the practical answer here: its amd64 packages are still built for the original x86-64 baseline. The old controllers keep working, and CERN does not have to rework boards, racks, drivers and the whole surrounding environment just to stay compatible with a new RHEL release.

There was a workaround, since AlmaLinux 10 has a build for x86-64-v2, but CERN chose Debian all the same.

The sudden death of classic CentOS, followed by restricted access to RHEL sources, almost certainly added to distrust of the ecosystem too. But it would be wrong to oversimplify: the deciding factor was neither ideology nor licence costs but compatibility with industrial hardware.

And yes, CERN is not “ripping out RHEL entirely”. The lab’s data centres and main computing stay on RHEL and AlmaLinux. What moves to Debian is specifically the accelerator control layer, where predictability and hardware lifespan matter more than trendy features.

The takeaway: don’t rebuild working infrastructure around a vendor’s whims; choose an OS that suits your hardware.

For the other side of depending on other people’s code, see bus factor = 1 in open source.

Sources

  1. CERN moves its accelerator controls to Debian — It's FOSS, 3 September 2026
  2. CERN's Debian migration moves 2,200 accelerator control machines off Red Hat — Hardware Busters, 2 September 2026
  3. Talk “Controlling CERN's accelerators with Debian” — MiniDebConf Winterthur, 30 August 2026
Your people nearbyAsk your own crowd — the chat answers faster than searchThe main chat of IT Party Pattaya: IT people, digital professionals and entrepreneurs who have already been through the same thing. Since July 2023. The chat is in Russian.Join the chat (in Russian)

Related reading