Comment by rmoriz | Hacker News Reader

Comment by rmoriz | Hacker News Reader

Hopefully this will drive many DE ISPs out of business. Strato, 1&1 and Hetzner are incapable of providing anything besides dumb webhosting and cheap bare metal stuff. I've wasted so many hours trying to explain these companies how the world is changing, they never got it and now it's too late to build anything that could compete with AWS.

They (1&1, Strato, HostEurope, Hetzner, …) don't even try to provide OpenStack packages or something leightweight-ish using Docker, except if you pay every single hour of their manual work. 1&1 build something called ProfitBricks.de which focusses of "designing your infrastructure in the browser", however they build a big Java EE legacy framework with a nasty SOAP-API. Their sales pitch is something like "we are cheaper than amazon/they rip of their customers".

And still they neither contributed support to LibCloud or fog.io so neither Chef, Puppet, Ansible nor SaltStack work out of the box with it. I don't understand those people.

I hope Hetzner will be able to compete, because their offers have so much better performance per EUR than Amazon's: https://www.hetzner.de/en/hosting/produktmatrix/rootserver-p...

I talked to some colleagues about scaling with Hetzner and their answer was to just overprovision like crazy since Hetzner's so cheap anyhow.

how? You don't have a shared storage nor a private network nor "availability zones". Sure, you can tinker something if you have free time but you'll move off and regret it. Like Wooga and others.

If you run some cluster FS on your single, external network interface, it's quite easy to DoS your cluster.

There is no option to place a service inside a DC, you can ask them but this will result in a manual provision taking days —or be ignored. Also you don't know how the different DCs are connected in terms of internal bandwidth and external reliability. Hetzner itself says their inter-dc connections "are not optimized", whatever that means.

If you run multiple customers on your system and one get's DoSed, Hetzner will disconnect your system. Sure, this was good enough 2-3 years ago, but they didn't change their DoS, provisioning and networking setup.

You don't need anything that you mention to scale. My company using hetzner for many years to build profitable business (not extremely big, but not "private blog") and we are happy with our choice. It probably took a slightly more time to build scalable solution with hetzner, but our monthly bill is 9 times smaller with hetzner (than comparable hardware with AWS). This savings allow us to hire 2-3 people more that easily covers additional work.

How do you define scaling if a manual process controlled by another party is involved? Besides that, what do you think that Hetzner does not provide a 2 factor authentication? As Hetzner doesn't use HSTS on their client/api interface, it's quite easy to run ssl downgrade attacks, sniff your credentials and format all your servers, delete all your backups.

I think you're confusing elasticity with scalability. Just because you can't spin up instances within seconds doesn't mean you can't run at scale on traditional hosting providers.

Still, I do believe that's the direction the industry will move towards. OVH, for example, is already pushing pretty hard in that market, and their business model was pretty similar to the providers you mentioned not so long ago. It's probably just a matter of time.

We don't have sudden huge traffic spikes and so we don't need automatic scaling. We maintain a few extra serves for redundancy and small traffic spikes. When business grows and we notice that current servers isn't enough we simply order a new ones.

> what do you think that Hetzner does not provide a 2 factor authentication

It's ok for me. Easy solution is to run VPN on hetzner server and access Robot (their control panel) only inside their own network.

> How do you define scaling if a manual process controlled by another party is involved?

Who even says you have to use one hosting provider? ;)

I think the problem here is you aren't used to regularly stable workloads that can be anticipated with reasonable lead time. [e.g. a week or more]

While they should implement HSTS, it's not really a problem for informed users; one can get the same protection by installing HTTPS Everywhere, which includes a rule for Hetzner.

Shared storage? Use Riak CS, MogileFS, ceph, LeoFS, Openstack Swift etc. as a S3 replacement.

Availability zones: Hetzner has multiple independent DCs and they are happy to place your servers in different DCs if you ask them. What Amazon calls "Elastic IP" they call "Failover IP".

Yes if you only need to host your blog yet you don't want it to go down when the HW inevitably fails, don't go to Hetzner because setting up Postgres failover and Chef and all that is complicated.

And if you're big / have enough money Hetzner's off-the-shelf HW and their unwillingness to offer an SLA will make you move away.

But there's a sweet spot for Hetzner, it's where you want good performance for a great price, have enough money/knowledge to setup proper bare metal deployments with failover yet don't yet need the guarantees of a "premium" DC or the features of AWS.

I know quite a few companies which fit that description.

Dude, chill out. I don't see how the fact that these providers don't target your specific needs imply that they need to be driven out of business.

I use both AWS and Hetzner. Both have very different offerings aimed at different use cases. AWS gives you simplicity, but everything is crazy expensive. Hetzner and friends give you much more powerful hardware, but you have to setup things yourself.

I have to setup things myself on AWS, too, but I can use standard tools like Chef or Puppet and don't have to deal with manual service requests or running a mash-VPN myself to connect my physical machines over a single network just to have some "private" networking.

I think DigitalOcean is a nice example of how fast you iterate. They did many things right, some wrong. But the don't settle. They are building new things like private networking, their v2 api, IPv6 (hetzner doesn't provide v6 on vservers iirc EDIT: they do now).

The default answer I got when asking for features that e.g. US, UK or FR based ISPs already have is: "But it's not hosted in Germany! The cloud is insecure!"

Today we know, that it doesn't matter, where in Europe you're located as intelligence services cooperate and share all data. So Hetzner for example is required to provide access to german intelligence services by law. They share it with the NSA. So it's FUD.

So maybe this provides some insight of my kind of emotional initial posting.

Hetzner does provide ipv6 on VPS. You just request it in the control panel and get a /64 allocated.

Hetzner can give you extremely cheap physical hardware, prices which you cannot hope to approach with AWS.

You don't need any of the buzzword-y technologies you list to be supported directly by the host. If you want Docker, you can run it yourself.

VMs are over-rated.

If you want to build something sustainable, you earn money or need to be able to scale incredibly well. If you just want to host your private blog, even DigitalOcean provides better packages than Hetzner.

I don't have a horse in this race, but IMO the point is more about AWS-provided infrastructure that lets people scale more easily than otherwise possible; things like Elastic Map Reduce, S3, DynamoDB, SQS + SNS, RDS, etc are less-easily replaced by home-grown equivalents. There are great open-source solutions for every one of those things, but the point is having to not manage them.

If you're using the cloud as "just another data center", I feel like you're missing many of the benefits of software-defined infrastructure and disposable systems.

exactly. Speed of innovation is how you win. If you start with an MVP, a cheap 5$ droplet at digital ocean is cheaper than the cheapest hetzner server. if you grow and need failover-solutions, you'll run into a lot of devops work when you build everything your own. (you can't automate a lot of work because of lack of API and automation by the typical providers)

So it's not a price question imho. It's about speed of innovation, usability of "building blocks" and limiting of operational risks.

Tying yourself to a specific provider is dangerous.

RDS and EMR are fine, since you can reproduce the exact same API anywhere else. S3 is also fine, since its API is simple enough and even reproduced by other vendors. Something like DynamoDB I wouldn't touch.

Things like AWS are great for when you don't yet know what you need, for absorbing load peaks and for getting something running very quickly. They are however the most expensive thing you can buy for what you get.

How would they go around selling something like docker? Selling servers with docker preinstalled? Why can't you install it yourself?

Because provisioning is slow, especially for additional IPv4 networks. They could provide a PaaS-Services like e.g. https://stackdock.com/ utilizing Docker. Or at least some virtual private-networking and iSCSI-block storage and object storage (like S3).

Honestly their cheap-server-design could be a great foundation for something, if they federate it and provide it. It's nearly unmanagble to do it yourself as you don't have a privat interconnect, availability/redundancy (e.g. in which DC the put a system) and you can only have one failover IP/network per system.

Also they have really bad luck with hard drives. I've had SMART errors/disk replacement in 3 or 4 disks on a single machine over the past 2 years.

Renting out docker containers on a server shared by multiple customers is a completely ridiculous idea from a security point of view though.

Reply on news.ycombinator.com