The epitaph of the switch: How CSC helped Finland’s science avoid a very large cloud bill

When research infrastructure works as intended, it enables science without calling attention to itself. This is a story on how CSC’s infrastructure quietly saved Finnish science millions by moving massive volumes of biomedical data securely and freely, at a fraction of the cost a commercial operation would impose on society.

During spring 2026, a CSC network expert will walk into the CSC data center, look at a rack, and conclude that a device called csc-bmi1-sw switch, which has been there since 2011, no longer needs to be there. Then he will unplug it. A small piece of Finland’s digital research history quietly disappears.

The story started about 15 years ago with a challenge. At the FIMM – Institute of Molecular Medicine Finland, sequencers and microscopes were producing more precision medicine research data about the Finnish population than anyone had planned for. Analyses were getting heavier. Storage filled up faster than budgets refreshed. Local IT at the University of Helsinki did their best as always. But the growing research organisation realized that information technology capacity was no longer solvable locally.

At the same time, CSC and Funet were building new computing services in the emerging ELIXIR infrastructure. We didn’t call it “sovereign European cloud” or “AI-ready infrastructure”. It was just a project, Biomedinfra.

A simple idea with infrastructure-scale impact

The idea was simple. What if we didn’t move the computing from a lab to CSC, but quietly extended CSC compute to the lab? In cloud terms: What if we provided our infrastructure as a service to a university?

To make that idea real, we had to make a practical decision: installing an optical private network between FIMM and CSC, including buying a dedicated, then high-end 10 Gbit/s network switch called csc-bmi1-sw. At the time, it felt like just another component procurement in the Funet national network, one decision CSC often makes.

That switch, however, did something subtle and powerful for national supercomputing: it made distance irrelevant thanks to the high performance of the network. Suddenly, scalable computing capacity wasn’t “over there at CSC” anymore. It was available nationally inside the University of Helsinki network.

Once the switch was in place, infrastructure changed for the local researchers at FIMM, but not in a dramatic way. The local computing cluster just started working better. Biological data started to move between CSC and FIMM at scale. Computing jobs could also scale inside the local system where sensitive data resides without logically moving data away from the organisation. Workflows didn’t need to be redesigned and moved every time a research problem got bigger. Sensitive biomedical data stayed under control of the organisation responsible for it, but analysis capability was no longer constrained by local limits.

Some researchers at FIMM at that time did not even think about the change or the switch at all. This is about the highest compliment you can receive when operating research infrastructure services.

Over the years, the traffic grew and grew. Looking back, we estimate that the switch has carried something like:

5 Gbit/s x 60 seconds x 60 minutes x 24 hours x 365 days x 15 years
= 2 365 200 000 gigabits of data

It is hard to fathom what this means. Inside the Finnish research infrastructure, this didn’t trigger alarms or, most importantly, invoices. Data moved because research needed it to move as part of supercomputing to solve precision medicine problems at FIMM.

A public infrastructure cost model that just works

The CSC cost model was simple: build it once, use it freely.

The switch itself cost somewhere around €60k–€110k, was paid by the Research Council of Finland grant for ELIXIR, and lasted 15 years. It ran the last seven years without interruption. The switch rarely required any attention from the Funet expert crew overseeing it. It never caused a hassle, it just sat there blinking in the CSC data center.

Now, let’s imagine a different version of events.

Same molecular medicine research, same data. But if, instead of CSC, the cloud computing service that FIMM uses would live on, let’s say, Amazon Web Services or Microsoft Azure.

Commercial cloud providers like Amazon and Microsoft typically charge no fee for data coming in (ingress) but base their business model on the outbound data transfer (egress). Outgoing data transfer cost at Amazon or Azure is typically in the range of €0.05–0.09/GB depending on scale and connectivity.

Computing related to precision medicine over 15 years, reaching approximately 300 PB of data returning from the cloud over time, would cost around €15M at this rate. That is about one million euros per year to move photons that carry data – not counting precision medicine expert time, computer GPU/CPU consumption, or storage space.

No one would actually run research infrastructure services at this cost. You would optimize, minimize, compress, avoid transfers, redesign workflows. 

In this version of the story, organisations would start thinking: Do we really need to move this data? Can we afford to rerun our analysis? What happens if we don’t carry out research at all?

In the true version of the story, CSC and the switch removed those questions. This is why CSC exists in the Finnish research ecosystem: to support data-intensive computing needs, like AI/ML model training, using high-quality data.

We focus on enabling science at a fraction of cost that commercial operation would cost to society. The technical system disappears into the background when it works. Without the CSC infrastructure as a whole, a lot of precision medicine research at FIMM in the last 15 years might not have been carried out.

The last seven years of the switch’s life are perhaps the most impressive: Nothing happened. No outages, no incidents, no surprises. In infrastructure terms, silence is success.

15 years later: Building public research infrastructure at European scale

The workloads for precision medicine research have moved on, and CSC architecture has evolved. The Funet network has been scaled beyond a single connection and is more resilient. CSC prepares for next generations of computing hardware like Finland’s new national supercomputer Roihu in summer 2026, and the upcoming LUMI-AI supercomputer and LUMI-IQ quantum computer arriving in 2027-28.

And the switch? It has done its job and can be turned off.

This is a small story of a small device, but the same pattern is now playing out across Europe. Through infrastructures like ELIXIR and initiatives around health data, AI, and high-performance computing, we are asking the same question again: How do we let data move freely without losing control or slowing science down?

The scale is larger now, but the core idea for building public research infrastructure is the same: connect distributed systems on the Internet, keep data protected by design, and don’t charge for every data movement operation.

When the switch is finally unplugged, nothing dramatic will happen. Research at FIMM will continue, and data will still keep moving on the CSC ePouta cloud with upgraded network technology. This is the perfect ending for the switch. The best infrastructure is when it becomes so reliable and invisible that we forget it was ever there at all.

This article was first published on 12 May by CSC.