CERN is moving 2,200+ accelerator control computers to Debian 13 by the end of 2026. That announcement prompted Mohamed Elashri to revisit a decision from 2019: Fermilab and CERN ending Scientific Linux.

His claim is not that Scientific Linux was technically better, and not that the labs could have controlled what Red Hat did afterwards. It is that the retirement priced only the visible cost of maintenance and ignored what the capability was worth.

Scientific Linux grew out of a real problem. High-energy physics spans labs, universities and countries, and divergent glibc and packaging environments break collaboration fast. The community needed something stable for years, freely redistributable, enterprise-compatible, and licensable for every machine without a commercial key.

  • Fermilab announced it in 2003; CERN joined soon after; it spread to universities, companies, and systems aboard the ISS.
  • The value was that the scientific community could reproduce part of the Enterprise Linux platform itself, rather than only consume it.

After Red Hat and CentOS joined forces in 2014, consolidating on CentOS looked strictly rational. Why spend scarce engineering time duplicating a rebuild someone else already maintained, backed by the vendor itself? In 2019 Fermilab concluded there would be no Scientific Linux 8.

What the labs gave up was a credible exit. The cost of maintenance was visible; the benefit of independence was not, because nothing had failed yet. That is how redundancy always looks before it is needed.

Then the assumptions broke in three stages:

  • December 2020: Red Hat replaced CentOS Linux with CentOS Stream, which sits ahead of RHEL rather than behind it. The product everyone had standardized on effectively ceased to exist.
  • 2022: CERN and Fermilab recommended AlmaLinux — another independently governed RHEL-compatible rebuild, the exact category they had just retired.
  • 2023: Red Hat changed how RHEL sources were published, making the rebuild ecosystem’s dependency structure much more visible.

Technical compatibility with an upstream is not independence from whoever controls it. The Debian trigger was less ideological than that: RHEL 9 raised its x86-64 baseline and RHEL 10 raised it again. Accelerator controls interface with custom electronics, legacy buses and real-time systems where replacing one computer can mean redesigning the boards attached to it. A 2023 risk analysis put staying on the Red Hat ecosystem at roughly 5.4 million CHF, plus redesigning around eleven boards and recommissioning equipment. They replaced the operating system instead.

Elashri is careful about the counterfactual. IBM closed the Red Hat acquisition in July 2019, after Fermilab had already announced there would be no Scientific Linux 8, so this is not a story about IBM killing something foreseeable. His narrower claim is that independent alternatives change incentives: a vendor behaves differently when major customers have a credible exit. Competition matters even when nobody switches, because the possibility of switching matters.

The generalizable part is the retirement checklist. Before retiring institutional infrastructure, account for governance, concentration risk, migration cost if an upstream changes direction, the institutional knowledge being discarded, and the value of an alternative you hope never to need.

We are good at calculating the cost of maintaining something ourselves, and bad at calculating the cost of losing the ability to maintain it. As CERN’s own accelerator team put it: distribution portability is a very good thing.