Comment by ThePhysicist | Hacker News Reader

Comment by ThePhysicist | Hacker News Reader

Parent Full thread

We host a lot of our infrastructure on Hetzner Cloud and are very happy, but to be fair it's a long long shot away from what AWS offers. You basically only have compute, storage and networking, only three datacenters and two regions to choose from and no managed services at all. You also can't do advanced networking like announcing your own IP addresses (Hetzner only allows that in colocation). You also only have a few instance types available, with RAM maxing out at 32 GB and no GPU servers available.

Don't get me wrong I love Hetzner and their cloud offering and I think they're doing a fantastic job, but it is not comparable to platforms like AWS, Azure or GCP. It's good enough for simple use cases but I think most large companies that want to switch from on-premise would have a hard time adopting it since so many crucial features are missing.

The LIDL cloud will have the same problem I think: They will probably build it on top of OpenStack or Kubernetes but will never reach feature parity with AWS, Azure or GCP. I think in their niche (retailing, logistics) they might be able to get some good adoption if they offer specific services and infrastructure based on their own use cases and experience, but this isn't really competition for AWS, at least not in the broader sense.

Honest question, Kubernets does not helps in closing the gap between AWS and Hetzner Cloud?

I would think that an honest setup supported by k8s would, at least alleviate the need of managed solution?

Are there AWS services without a competitive open source solution deployable in k8s?

Kube removes some of the infrastructure burden of running your own solution but not all of it and that's not the only burden.

For example, I can deploy Kafka on k8s. I still need to know enough about Kafka to be able to deploy it correctly (zookeeper, etc.). I need to know enough about k8s to know how Kafka can run on i and to fix issues. I then need to know enough about Kafka itself to tune it, fix issues, manage it, etc. Then there's interaction events such as upgrades which require coordinating between k8s and Kafka. And in the end I'd still be a novice at all of it.

I would guess that we could just give an external company control over a kube master and have them run their operator, which in turn will spin up all the necessary pods and stuff.

You give a SSH certificate and you get back an IP address in your on-premises / cheap host with Kafka running along with monitors.

Serverless? K8s is explicitly not operating in the same space as triggers and code not bound to any real instances.

https://github.com/kubeless/kubeless seems pretty healthy to me

> RAM maxing out at 32 GB

This part is just not accurate. I'm running their dedicated server with 256 GiB RAM right now.

Looking at https://www.hetzner.com/cloud, the top of the line dedicated option seems to be CCX51, which only has 32GB (edit: 128GB, I must have misread). Do you have some sort of special deal with them?

Me and the parent commenter are talking about the Hetzner cloud offering, where the largest instances have 32 GB of RAM (https://www.hetzner.de/cloud). Sure you can get a dedicated server there with much more RAM and you can host your own hardware in their colocation facilities, but that has nothing to do with cloud computing in the common understanding of the term.

He's taling about the cloud servers, they are indeed capped at 32GB. You're talking root servers. https://www.hetzner.com/cloud

For their dedicated servers you can have at least 712GB RAM.

But for their cloud offering the maximum is only 128GB RAM.

See https://www.hetzner.com/cloud and click on "dedicated vCPU". btw: do not confuse "dedicated vCPU" with dedicated server.

I wasn't aware of that offering, thanks for letting me know.

Hetzner is great until you have some sort of issue.

They seem to either not have much in the way of internal procedures or just ignore them if they exist.

As long as you're prepared for all of your data and backups to just disappear, it's a great low-cost environment.

If you expect their operational team to not accidentally delete your environment, it just hasn't happened to you yet.

> They will probably build it on top of OpenStack or Kubernetes They are building it on top of OpenStack you can see it in there job offerings [1]

[1] https://stackit.de/de/karriere/

Reply on news.ycombinator.com