The Cost of Cloud, a Trillion Dollar Paradox

I think the problem is using "the cloud" eroded knowledge of physical infrastructure in new DevOps guys.

Going to the DC once every few months to add new servers was a good way to remember how to deal with hardware and train new people. I used to take interns/junior devs just to show them how it looks.

Theory: The cloud took over the same way most trendy tech took over: resume driven development. Everyone on all the teams pushed for it because of how much value it adds to their resume and gradually more and more companies bought in, which accelerated the process because company d started feeling worried that they were behind the times since companies a-c already switched

The cloud is great for startups and tiny companies though

Very wordy article but interesting point about how cloud is considerably more expensive than on-prem at a large enough scale.

I don't believe the general user gets much benefit from using a general purpose "everything cloud". The mere scale of building and maintaining such a cloud lead to both these bills but also makes them so much more complicated.

I believe we will see more purpose built clouds that target specific use-cases and infra paradigms.. or at least I hope so as I'm part of building such a service.

Division of labor and specialization of trade drive it as an optimization of collective deference to a pool of specialized talent at as large scale as possible. There is no need and only excess cost for every company to run 10k or 200k servers themselves more-or-less exactly the same as anyone else would when they can have someone else deduplicate that effort. For example, I work with endpoints instead of servers. That's not seen clearly as a value-add enabler as much as a necessary business expense activity. Gathering larger and larger numbers of similar users is cheaper when the problems are similar enough to be solved fewer times, preferably by someone else and for less money. It's only when the providers charge too much or cannot provide nuanced-, attuned-enough service that customers look to move. It's a utility, like email, that hardly changes once established because the costs of moving can be painful, time-consuming, and expensive.

This is a really flawed analysis, it ignores things like the cost and disruption of transition (both internally and to customers), the constant R&D needed to continually improve the compute base hardware and software architecture, the people needed to do all this and their skills (which need constant improvement) and availability (the best people are not easily replaceable units of work and it takes time for new people to come up to speed), security, and more. It basically only looks at _estimated_ costs based on a very few cases. Many big software projects fail, and moving from cloud to in-premises (and vice-versa) is a big software and hardware project. Some of these factors are pro-cloud, some are anti-cloud depending on circumstances. Cost should not be the only determinant.

My personal opinion is that cloud is currently a major security vulnerability. In a time where computer security is very obviously not a solved a problem, betting on a monoculture (similar to mono-crops in agriculture) means that if one major attack is successful then everything that depends on that cloud architecture is at risk (potato famine). So far we've been fairly lucky (as far as we know; it could be that compromises exist and are being exploited without doing anything so obvious as crashing the cloud). But all the eggs are typically in one basket. If the basket falls all the eggs break. On premise data centers can be compromised in the same way, but it's more work for the attacker and less profit, raising the barrier. Until we figure out how to truly secure compute infrastructure, security through obscurity (strangely enough) and diversity, by having every company build their own data center, seems to be better for the security of society overall (though any one company can still be compromised.) In part, it's the old distributed versus centralized argument. A highly variable distributed set of systems is more robust against attack than the monoculture and single point of failure of a centralized system, though a centralized system offers more control (and wow, CEO's really like centralized control for obvious reasons! It's a major flaw in their outlook, similar to why the Russian army is having problems because of their rigid top-down hierarchy.) Look to nature for examples of what survives. Insects are considered likely to be the last survivors on Earth, dinosaurs are (mostly) gone. /rant

I have a hunch that the chief issue with the cloud is that it enables and promotes easy sprawl of the infrastructure.

It's just so easy to "scale" your way out of glaring performance bugs such as unoptimized queries where that becomes the default behaviour of most engineering departments as horizontal scaling is often already baked into the different cloud services.

On the other hand, hunting for the root cause of performance issues is a complex, uncertain process especially with the crazy Rube Goldberg contraptions that most cloud solutions tend to mutate into.

"...AWS still operates at a roughly 30% blended operating margin net of these discounts and an aggressive R&D budget — implying that potential company savings due to repatriation are larger"

to suggest individual companies can 'optimize' to a cloud corps utilization (via aggregation of demand) and hence reap the 'savings' is pretty dishonest.

This is being sold as a value gain in the article but surely the hyper scalers game a corresponding hit. And everyone collectively takes a loss given ineffificiencies of disaggregating this

I thought this looked familiar: the title should have (2021). The dramatically increased cost of capital presumably makes this proposal less attractive.

Is repatriation lingua franca within cloud executive circles?