We were wrong about GPUs

Then you seem to be confused, because you're saying Kubernetes but what you're actually talking about is implementing a toy container orchestrator.

This (and a few similar upthread comments) sum the problem up really concisely and nicely: pervasive, cross-stack understanding of how things actually work and why A in layer 3 has a ripple effect on B in layer 9 has become increasingly rare, and those who do know it are the true unicorns in the modern world.

Big part of the problem is the lack of succession / continuity at the university level. I have been closely working with very bright, fresh graduates/interns (data science, AI/ML, software engineering – a wide selection of very different specialisations) in the last few years, and I have even hired a few of them due to being that good.

Talking to them has given me interesting insights into what and how universities teach today. My own conclusion is that the reputable universities teach very well, but what they teach to is highly compartmentalised and typically there is little to no intersection across areas of study (unless the prospective student hits the pot of luck and enrolls in elective studies that go across the areas of knowledge). For example, students who study game programming (yes, it is a thing) do not get taught the CPU architectures or low-level programming in assembly; they have no idea what a pointer is. Freshly graduated software engineers have no idea what a netmask is and how it helps in reading a routing table; they do not know what a route is, either.

So modern ways of teaching are one problem. The second (and I think a big one) is the problem that the computing hardware has become heavily commoditised and appliance-like, in general. Yes, there are a select few who still assemble their own racks of PC servers at home or tinker with Raspberry Pi and other trinkets, but it is no longer an en masse experience. Gone are the days when signing up with an ISP also required building your own network at home. This had an important side effect of acquiring the cross-stack knowledge, which can only be gained today by willingfully taking up a dedicated uni course.

With all of that disappearing into oblivion, the worrying question that I have is: who is going to support all this «low level» stuff in a matter of 20 years without a clear plan for the cross-stack knowledge to succeed the current (and the last?) generation of unicorns?

So those who are drumming up the flexibility of k8s and alike miss out on one important aspect: with the lack of cross-stack knowledge succession, k8s is a risk for any mid- to large-sized organisation due to being heavily reliant on the unicorns and rockstar DevOps engineers who are few and far between. It is much easier to palm the infrastructure off to a cloud platform where supporting it will become someone else's headache whenever there is a problem. But the cloud infrastructure usually just works.

And this is whom I ended up hiring – young engineers with curious minds, who are willing to self-learn and are continuously engaged in the self-learning process. I also continuously suggest interesting, prospective, and relevant new things to take a look into, and they seem to be very happy to go away, pick the subject of study apart, and, if they find it useful, incorporate it into their daily work. We have also made a deal with each other that they can ask me absolutely any question, and I will explain and/or give them further directions of where to go next. So far, such an approach has worked very well – they get to learn arcane (it is arcane today, anyway) stuff from me, they get full autonomy, they learn how to make their own informed decisions, and I get a chance to share and disseminate the vasts of knowledge I have accumulated over the years.

> How can one claim to have a B.Sc. in our field without understanding how […]

Because of how universities are run today. A modern uni is a commercial enterprise, with its own CEO, COO, CO. They rely on revenue streams (a previously unheard-of concept for a university), they rely on financial forecasts, and, most important of all, they have to turn profits. So, a modern university is basically a slot machine – outcomes to yield depend entirely on how much cash one is willing to feed it. And, because of that, there is no incentive to teach across the areas of study as it does not yield higher profits or is a net negative.

This is why numbers do not stack up in the calculations – the premise that the DB has to be provisioned is not the correct one to start off with.

The right way of cooking RDS in AWS is to go serverless from the start and configure the number of DCU's, e.g. 1 to N. That way it will be even cheaper than the originally quoted $200.

Generally speaking, there is absolutely no need for anything to be provisioned at a fixed compute capacity in AWS unless there is a very specific or an edge case that, likewise, warrants a provisioned instance of something.