[Lorenz-Kraft]

We made the experience that its very cheap and easy to do things on barebone metal by yourself IF you have the knowledge in-house anyways. By cheap, I mean 10% of the costs. You might add another 10% costs to have an external backup service running (for example the Backblaze B2B).

[akcura]

Thanks Lorenz. Did you have any reliability issue on the machines? We need them to be up and reliable. As you said, the cost seems to be one fifth.

[Lorenz-Kraft]

We were using simple Hetzner 49Euro/month (SSD/NVME) Servers of the shelf (these do not include Terabyte Storage, this will be a paid upgrade (10 Euro/m ??)). The "reliability" of the servers are normal. So once every two years or so (my personal experience, might be worse, might be better) there are hiccups with ventilators or something. So if your tech knowledge is allready in-house, you should put a little effort into fallback or load balancing to not disturb your daily business. Also, a automate spinup of servers (fixed install and setup routine) should be your best practice.

Bye the way, your in-house tech (and proggers) should keep an eye on "Vendor Lock in". As soon as your are using services that are not open source, you will tie yourself to one company and its pricing scheme. Getting out of this rabbit hole is mostly painful.

Depending on your analytics needs, you might even consider having your analytics server IN YOUR office. For example, if you are doing any kind of deep learning, you can buy "little" servers and two or three top notch GPU (Graphic Cards) per server to fit your needs.

Like I said, this all makes sense if you allready have tech knowledge in-house. If you don't then I would strongly advise to use providers like AWS and such, because they are handling all the above described. Also, if your data access is critical to be available 100%, these providers often give you pre build solutions that are battle proven.

[akcura]

Thanks!