Cloud and skills, part 2

This article follows on from a previous one, in which I talked about building individual skills, taking as an example a developer who deploys an application on PaaS rather than deploying it on a server, at least once. In this post, I mean to let that line of thought drift toward how skills get built at the scale of our country.

Watch out, this is going to go far!

Hari Seldon in da house

Among the dozens (hundreds?) of good ideas you find in Isaac Asimov’s books, there’s one that has been nagging at me for more than 20 years (yes, that’s a lot).

In his “Foundation” cycle, a Galactic Empire, ruling the universe, collapses. To make it possible to rebuild civilization, the whole of humanity’s knowledge is gathered on one planet, Foundation. Its people will have the job of rebuilding a civilization after the 10 centuries of dark ages that are forecast.

So, no, I’m not going to talk about the brilliant idea of “psychohistory”, the science which, in the novel, lets you predict the future from past and present data — hello machine learning. No, what interests me here is the loss of knowledge and of skills.

In the book, atomic energy goes, in the space of 1 century, from a science that scientists master, understand, and run, to a kind of magic, even a cult, where the few technicians who still keep that energy source running are seen as priests. Humanity has lost this knowledge, and fast.

This is only science fiction. Fortunately, in reality, it can’t happen. For example, we know precisely how the pyramids were built (#joke).

Can this still happen to us, at the scale of humanity? Of a country? Of an individual? These are obviously purely rhetorical questions. Our modern civilization is the result of hundreds of years of acquiring and sharing knowledge. If we have indeed lost the know-how behind a lot of techniques, it’s often for a good reason: a technique shows up to replace another, bringing some “better”: better efficiency, lower cost, better yield, better safety, etc. The old technique, for its part, gets dropped. And a technique that is dropped, and no longer practiced, is doomed to disappear.

Nuke them all

In my previous post, I wrote a slightly provocative sentence: “serverless” makes me think furiously of the “factoryless” of 30/40 years ago.

40 years ago, in France, politicians of every stripe dreamed of a France without factories. A factory is dirty, and besides there are a lot of poor people working inside, so no more factories, no more poor, magic. I barely exaggerate.

It was a great idea, applied to the letter at Alcatel (for example) by a few geniuses. And oddly enough, in 2022, we no longer know how to make telecom equipment in France (or very little / very specific), but that’s lovely, because we can buy it from our former Chinese suppliers! (I could talk about the Alcatel engineers in 2005 who mocked Huawei’s gear: “they even copied the layout of the mod op, those dumb*sses”…)

Could we launch a French telecom equipment maker to compete with Huawei in 2022? Do we still have the skills we need in R&D, in engineering, in production? And if yes, at what price?

Giscard Punk moment: in the 60s/70s, France built itself a real know-how, recognized worldwide, in producing nuclear energy. We were strong, we exported our know-how, and we innovated, on the energy source best suited to climate change: abundant, dispatchable, and very very low-carbon (a hot take on my part, but on this subject I run hot).

For 25 years we’ve been talking about STRONGLY cutting the share of nuclear in France. We stop the experimental programs, we shut plants, we no longer bring reactors online (the last one dates from 2002!).

What is the impact of those decisions on whether skills are still there in the nuclear sector? Never mind the highly qualified researchers and engineers: we can ask the question about maintenance technicians, welders. In 2022, while half the fleet is down because of maintenance defects, we can no longer find enough trained welders, and we bring them over from the US.

One last word on managing skills in nuclear, found thanks to Thierry Caillon. I recommend reading a report from 1998 (25 years ago!) from the Parliamentary Office for the Evaluation of Scientific and Technological Choices, and in particular chapter III on “The effects of not carrying out the EPR project for the manufacturers”.

[ … ] defining an energy policy has to be understood dynamically. That position implies keeping the skills of the nuclear industry, because the cost of reconstituting those skills would be prohibitive. So what do you do between 2000 and 2010 if the EPR is not built? Ten years of theoretical research will certainly keep nuclear-physics research at a high level, but industrial know-how, which is extremely precious, also implies sharp, specialized knowledge and a network of subcontractors and manufacturers held to quality standards that are not those of the rest of industry.

AWS, GCP & AZURE

What’s the link with computing?

The cloud is the natural evolution of computing infrastructure. It lets you pool physical resources by building layers of abstraction in software. That software automatically and dynamically handles the allocation of memory, cpu, disk space, and so on. The added value is substantial, and putting it in place takes a high level of skill: low-level development for the network problems, modifying OSes, building the tools that configure resources and manage them automatically, and so on.

In 2022, the cloud market is sewn up by 3 big international names (Amazon Web Services, Google Cloud Platform, Microsoft Azure) and by (many) smaller ones. In France we also have very good players: OVH, Scaleway, Clever-cloud, Outscale, etc., but nowhere near the catalog or the volume of business of the big 3.

Through these providers, a computer person can do everything: load that scales by itself, code that runs on demand, a 15-minute compute every night on an ultra-powerful server, handling authentication for an application, using an NLP model, and so on.

All of that, without ever touching a server! I haven’t found reliable numbers, but I’ll bet the number of sysadmins in the world must have dropped drastically over the space of 10 years.

For reasons of cost, of efficiency, of yield, the cloud is often the best choice there is. And given the steady stream of announcements from the big CAC groups handing their infra to AWS for some, to GCP or Azure for others, the trend is speeding up. On top of that, these big projects are sometimes given to joint ventures billed as sovereign cloud / trusted cloud: the offer still sits on the 3 big American cloud providers (Thalès x GCP, for example).

What I notice is that the names of the French cloud providers rarely come up, and that’s a real strategic problem if you look at it through the prism of skills.

When I write that “serverless” makes me think of the “factoryless” of 20 or 30 years ago, what I mean is that refusing to use French cloud services when it’s possible guarantees the loss of the sharp skills you need to build a cloud service, or at least their growing scarcity.

I’ll be told that the catalog of services the big 3 offer has nothing to do with what the European clouds offer. Yes, and to drive it home, a figure that makes you dizzy: in 2021, Amazon (not just AWS) spent $56,000,000,000 (yes, 56 billion USD) on R&D (for comparison, the CNRS budget is €3.82M for 2022). Their lead is considerable, and the battle may already be lost, with a France/Europe already out of history.

Still, we have the skills and the ideas to do quality cloud, with products that stand apart: for example the ReNESENS project, bringing together a cloud player (clever-cloud), an open-source virtualization vendor (Vates), a server designer (2CRSi), and a specialist in DPU processors (Kalray). A project and a consortium like that one has a lot more bite than a Thales/GoogleCloud, and it’s a lot more Giscard Punk 🤘.

We have gems, so let’s use them, because one thing is certain: if we don’t use their services, and without handing them €€€, there will be no lag left to catch up, since our French / European cloud industry will be dead, and it will take the skills we need down with it.

THE END.

The Giscard Punk visual comes from a subreddit of the same name, which I recommend you go look at. Past the light tone, the deindustrialization of France over 40 years will jump out at you.