Hacker News Reader

Comment by minkles | Hacker News Reader

First we got VMware because systems were getting larger than the load we could push onto them. Then we got Hyper-V because it was free and VMWare was getting needy and expensive. Then we moved to the cloud because it was "cheaper". Then we ran Kubernetes on a cloud node because it was hard to manage the workloads. Now we are having cloud cost difficulties.

Hang on a minute... deep thought ... surely we fucked up somewhere?

View on HN

Get off the cloud. Go with a reputable VPS provider like OVH or Hetzner. Get a machine allocated for a fraction of the cost of your cloud bill, and install dokku (or whatever) on it. Done.

Too late. Leveraged too much AWS stuff to escape.

Then it's burn out or golden parachute depending on your socioeconomic status.

Suck 'em dry and steal from the corpse.

I doubt it's too late. There may be a cost to escape, but it may not be insurmountable.

At the height of the cloud hype, we were all more than happy to pay to lift-and-shift our way to higher operating costs. Surely, then, a lift-and-shift toward lower operating costs is comparably feasible.

If anything the barriers are social. At the height of the cloud hype, we all believed we were riding unicorns to the end of a rainbow with a pot of gold waiting for us. When you believe that, of course you want to spare no expense on ensuring you have the fastest unicorn so you can get that pot of gold before anyone else can.

How do you hire people of you say “We have no K8s”?

It’s like applying to a job with a @hotmail.com address.

Seems like a good, self selecting filter for finding folks who share the same values.

There's several popular technologies where that I would regard a general, background antipathy towards as a positive hiring signal and Kubernetes would be one of them.

Why? It’s a decent open source orchestration software. Is there a better one to be using?

I think it's just one of those things that people tend to use when they actually don't need something that complex. It seems to have supplanted a lot of other tools in domains where you're not doing dynamic scaling or any of the other things it's really good at.

If you already know it then fine - use a hammer to crack a nut. But I get the feeling that it's the first port of call for people who probably suffer the learning curve without really getting anything much in return.

But I think another way to express my original point would be "Being an uncritical advocate for K8s positively correlates with having a tendency to overcomplicate things".

…you install K8s on your Hetzner machine?

OVH and Hetzner are not reputable providers. Not in this universe.

I agree with the use of Hetzner. I use them. Very cheap machine, very powerful. Very simple and straightforward. People need to think carefully about the possible expense and complexity of AWS. Hetzner is simple and straightforward.

Don't forget that this isn't an all or nothing idea. Sometimes it is (and should be), but there may be some things that are better/cheaper in the cloud, and some things that are better/cheaper done in a different way. Depending on how much of each you have it might be worth migrating the rest as well of course.

I’m struggling to think of a single thing that is better done at AWS’ 10x-100x inflates prices.

I work on a product with seasonal useage. If we did it on in house servers we would need 10x as many servers to handle the peak, but by putting it on AWS we pay for the servers only when we need them.

I doubt AWS is 10x more expensive when you consider all the costs - you have to pay for the humans to manage the servers if you do them in house.

bollocks. we have to pay for cloud people to manage the cloud stuff, and they're the same people except they have a cloud certification and can charge $8/hr more.

labor savings and simplification of processes with cloud never happened; if anything they got worse, and OpEx is higher than ever.

I have come to believe that, when weighting design factors that influence overall product development and maintenance cost, nowadays most people will choose an exponential curve over a logarithmic curve.

Because when you're just getting started, you're just looking at what's immediately in front of you. Logarithmic curves go up quickly in the beginning. Exponential curves are nice and easy in the beginning. The fact that one levels off and the other keeps turning ever more sharply upwards doesn't matter. Thinking ahead like that, considering second-order effects, etc., is neither Lean nor Agile, and invites accusations of engaging in Analysis Paralysis.

> because systems were getting larger than the load we could push onto them ... surely we fucked up somewhere?

Clearly the answer is to run everything bare-metal on a rackful of Raspberry Pis. ;)

What could go wrong?!

On the other hand, I don’t feel like this will be as amusing a suggestion for much longer.

I haven’t tried a Raspberry Pi 5 yet, but an Orange Pi 5 is pretty powerful.

I assume your point is that raspberry pi is a broadcom product

Run Kubernetes on Hetzner with something like Talos Linux.

Reply on news.ycombinator.com