A latest CERN announcement caught my attention. By the end of 2026, additional than 2,200 manufacturing computers and embedded systems about CERN's accelerator complex are expected to be operating Debian 13.
CERN is not abandoning the Red Hat ecosystem. AlmaLinux and RHEL remain the chief supported Linux distributions throughout much of the organization; Debian assistance is currently constricted to accelerator front-end systems.
Still, the move took me rear to a decision made additional than a decade ago: CERN and Fermilab's gradual abandonment of Scientific Linux. With hindsight, I think it was a mistake.
Not since Scientific Linux was technically better to CentOS, or since maintaining another Linux allocation was free. And not since CERN or Fermilab could somehow have controlled what Red Hat or IBM afterward chose to do.
The error was treating Scientific Linux chiefly as duplicated engineering work. It was infrastructure.
What Scientific Linux was solving
Scientific Linux grew out of a applicable issue in High Energy Physics. Large experiments extend laboratories, universities, computing centers, and countries. If one location builds against one type of glibc, another uses item slightly different, and a third runs an entirely distinct packaging environment, things rotate into achy extremely quickly.
The specialized community additionally needed an different combination: a Linux allocation that would remain stable for years, activity alongside endeavor software, be openly redistributable, and run throughout institutions without requiring a business licence for all machine.
Red Hat Enterprise Linux provided the stability and lengthy lifecycle, and Red Hat published the origin needed to rebuild it. Scientific Linux turned that origin into a community resource. Fermilab announced the allocation in 2003, and CERN joined shortly afterward. It eventually dispersed far beyond those two laboratories. Universities, investigation institutions, experiments, companies, and equal systems aboard the International Space Station used it or distributions derived from it.
Scientific Linux was never merely "RHEL alongside a distinct wallpaper." It gave the specialized community an institutionally autonomous implementation of the Enterprise Linux platform.
Then CentOS looked akin the apparent answer
When Red Hat and CentOS joined forces in 2014, moving to CentOS looked absolutely reasonable. CentOS already offered what many Scientific Linux users wanted: a openly accessible Enterprise Linux rebuild backed by a much larger general-purpose community.
CERN began moving its next important publish from Scientific Linux CERN to CERN CentOS 7. Scientific Linux 5 and 6 remained supported, but the forthcoming phase at CERN would be according to CentOS. The rationale was compelling. Why should CERN and Fermilab expend scarce engineering period rebuilding a Linux allocation whenever CentOS was already doing basically the identical work?
Why keep an HEP-specific allocation whenever the specialized community could converge on a larger average platform? If Red Hat itself was supporting the CentOS project, that seemed to create the phase additional sustainable, not less. In 2019, Fermilab took the disagreement to its logical decision or its literal "end." There would be no Scientific Linux 8. Fermilab would deploy CentOS 8 alternatively and activity alongside CERN and another laboratories to enhance CentOS for high-energy physics.
Taken on its own, this was not an irrational decision. But it removed item that was difficult to see on a spreadsheet: a dependable exit.
A allocation has choice value
This is what I think we underestimated.
The disbursal of maintaining Scientific Linux was visible. People had to rebuild packages, test updates, keep repositories, create releases, grip safety updates, and assistance users.
The advantage of an autonomous allocation was harder to quantify. As lengthy as CentOS behaved exactly as CERN and Fermilab expected, Scientific Linux looked redundant. That is how redundancy continually looks whenever nothing has unsuccessful yet.
Scientific Linux gave the specialized community its own implementation of a RHEL-compatible computing environment. More importantly, it kept living the people, processes, infrastructure, and organizational cognition needed to keep one.
That capability had choice value. You power not need to exercise the choice this year, or equal this decade. It becomes precious whenever the assumptions underneath your chief phase change. Those assumptions changed amazingly quickly.
[!NOTE]
CERN uses a akin disagreement to validate the FCC: the cognition required to build ample machines specified as accelerators and detectors is precious and must be preserved. If Europe does not build the next device following the LHC, that ability may be lost, the next ample device may be built in China, and Europe may endure its scientific guidance in the field. The identical disagreement applies to application infrastructure.
The CentOS assumption did not last
In December 2020, Red Hat changed the function of CentOS.
CentOS Linux, the traditional downstream rebuild of released RHEL versions was discontinued in favor of CentOS Stream, which sits onward of RHEL fairly than rearward it. This was additional than a alter in publish cadence.
Organizations had standardized on CentOS since they wanted a openly distributable approximation of the current RHEL release. The merchandise they had standardized on efficiently ceased to exist.
CERN and Fermilab immediately had to reconsider their Linux strategy. CERN noted that the shorter CentOS Stream lifecycle was incompatible alongside several specialized use cases. The laboratories evaluated Stream, RHEL licensing arrangements, and the emerging Enterprise Linux rebuilds. By 2022, CERN and Fermilab were recommending AlmaLinux as the norm allocation for experiments.
The circularity is difficult to miss.
Scientific Linux had been former partially since maintaining another RHEL rebuild seemed unnecessary during CentOS existed. A few years later, the community needed an autonomous RHEL-compatible allocation again, so CERN and Fermilab adopted another independently governed RHEL-compatible distribution.
AlmaLinux is a fine project. This is not a criticism of it. The necessity never disappeared; lone our implementation of it did.
Then the dirt moved again
In 2023, Red Hat changed how RHEL-related sources were publically distributed. CentOS Stream became the sole community repository for that origin material, replacing the former git.centos.org publish model.
This did not create RHEL "closed source," notwithstanding how the alter was sometimes described. That characterization would be inaccurate. But it made the dependency construction of the Enterprise Linux rebuild ecosystem much additional obvious.
By then, the instruction should have been familiar: specialized compatibility alongside an upstream merchandise is not the identical as independence from the institution that controls it.
Scientific Linux had strategic value that was never correctly accounted for. Its beingness meant CERN, Fermilab, and another institutions were not merely consumers of an ecosystem. Together, they could reproduce a crucial part of that ecosystem themselves.
Once discarded, that capability is much harder to recreate than to keep alive.
And now CERN is moving part of the accelerator complex to Debian
The latest division makes this former particularly interesting. CERN's accelerator controls collection archetypal tried to remain inside the Red Hat ecosystem. According to the display at MiniDebConf Winterthur 2026, the first scheme engaged CentOS Stream, alongside Debian prepared as a fallback because, in the presenters' words, they could not oversee new surprises.
Then came a additional bodily problem. RHEL 9 raised its x86-64 baseline to x86-64-v2. RHEL 10 moved to x86-64-v3.
Replacing old hardware may be inconvenient but manageable for an average server fleet. Accelerator controls are not an average server fleet.
These computers interface alongside tradition electronics, specialized boards, bequest buses, real-time systems, and equipment installed throughout a huge accelerator complex. The hardware can remain operational for decades since replacing one device may average redesigning the electronics connected to it.
CERN's 2023 hazard inspection estimated that staying in the Red Hat ecosystem could disbursal concerning 5.4 myriad CHF. It could additionally necessitate redesigning about eleven boards, hiring additional engineers and technicians, reorganizing racks, recabling systems, and recommissioning equipment.
The resolution was refreshingly simple: do not substitute millions of francs' value of functioning hardware to fulfill an functioning system's CPU baseline. Replace the functioning system.
CERN chose Debian. By the end of 2026, additional than 2,200 manufacturing computers and embedded systems about the accelerator complex are expected to run Debian 13.
The irony is difficult to miss
What interests me most is not that CERN chose Debian, but why Debian was valuable.
The accelerator squad specifically praised its continued assistance for older and small average architectures and its ability, as a community-led distribution, to vie alongside Red Hat.
CERN is additionally sponsoring Freexian to fortify Debian's long-term-support ecosystem. There is an crucial regulation here: whenever an institution depends on infrastructure supplied by a community project, contributing resources to keep that project autonomous and fit may disbursal small than consolidating everything about one vendor ecosystem in chase of short-term efficiency.
That appears remarkably near to what Scientific Linux used to provide.
Scientific Linux was maintained by a amazingly small group
This is another logic I inquiry whether ending Scientific Linux produced the funds we assumed. The allocation was not maintained by hundreds of engineers.
Its project former lists lone a fistful of important developers in its afterward years, alongside Fermilab providing the chief sponsorship, build infrastructure, bandwidth, and website. That does not create the activity trivial. Release engineering, safety updates, testing, bundle rebuilds, infrastructure, and person assistance all obtain genuine time.
But those expenses have to be compared alongside the value they produced. Scientific Linux provided a stable computing phase to an enormous specialized ecosystem for approximately two decades. Preserving that capability should not have been evaluated solely in conditions of the engineer-hours saved by adopting CentOS.
The applicable inquiry was: what would it disbursal to endure the capability to run independently if the surrounding ecosystem changed?
The former decade gives us at smallest a partial answer.
Did Scientific Linux keep Red Hat honest?
There is a tempting, stronger argument: that Scientific Linux and another autonomous rebuilds constrained Red Hat, and that removing among the important alternatives helped concentration power about CentOS and RHEL.
I think there is item to it, but I would not current it as historic fact. We cannot cognize the counterfactual.
Scientific Linux likely would not have prevented Red Hat from changing CentOS. IBM completed its acquisition of Red Hat in July 2019, following Fermilab had already announced that there would be no Scientific Linux 8. It would hence be misleading to decrease this to a narrative concerning IBM slaying item CERN and Fermilab should have anticipated.
What I do accept is that autonomous alternatives alter incentives. A vendor behaves alternatively whenever important customers and institutions have a dependable exit than whenever leaving requires them to rebuild years of infrastructure.
This is not particular to Red Hat. It is essential dependency management. Competition matters equal whenever nobody switches; the possible of switching matters too. Scientific Linux provided that possibility.
And since important specialized institutions maintained it fairly than a business Linux vendor, its incentives were unusually fine aligned alongside long-lived specialized infrastructure.
The genuine error was confusing standardization alongside dependency
Standardization is good.
Particle discipline could not run efficiently if all university and lab invented its own incompatible computing platform. CERN and Fermilab were correct to desire average interfaces, compatible binaries, average packaging, and predictable functioning environments. But standardization and monoculture are not the identical thing.
The scheme went incorrect whenever we assumed that standardization required abandoning an autonomous implementation of the standard. Scientific Linux could have remained boring.
In fact, boring was exactly what we needed from it. It did not need to vie alongside Fedora on innovation or Ubuntu on desktop adoption. It needed to remain a reproducible, institutionally controlled Enterprise Linux platform: location the specialized community could go if the business ecosystem moved in a direction incompatible alongside specialized computing.
That is precious infrastructure equal whenever nearly nobody notices it. Especially then.
The instruction is larger than Scientific Linux
I do not think the answer today is to resurrect Scientific Linux. The ecosystem has moved on. AlmaLinux exists. Rocky Linux exists. Debian is proving increasingly helpful at CERN. Containers and contemporary application allocation have reduced the importance of the presenter functioning scheme for many workloads.
The instruction I obtain from Scientific Linux is concerning how investigation institutions value infrastructure. We are fine at calculating the contiguous disbursal of maintaining item ourselves. We are much worse at calculating the long-term disbursal of losing the capability to keep it.
When deciding whether to quit organizational open-source infrastructure, the calculation should contain additional than care hours. It should document for governance, concentration risk, the disbursal of immigration if an upstream project changes direction, the organizational cognition being discarded, and the value of a dependable substitute that you anticipation never to need.
The CERN accelerator team's final direction in its Debian display is likely the finest summary:
"Distribution portability is a extremely fine thing."
It took additional than a decade, multiple Enterprise Linux scheme changes, a new community rebuild, and now the immigration of thousands of accelerator computers to attain there. Scientific Linux was already instruction us that instruction twenty years ago.
We fair stopped listening.
