works for me in the baltics.

Part of the team here - Happy to discuss this further if you have any question about the move.

How much time did it take from start to finish in terms of planning and preparing the entire operation?

The actual migration, a day or less. Now there was a 2-week sprint for testing and validating backup and restore before doing the actual migration. At the non-engineering level (management and such) of course there was a review and consideration time as well.

Good question.

I'd also like to hear more detail (budget, time, people and skills) on real cases.

When I had research students I set a few of them doing studies on migration, degoogling, data "repatriation" (ick!) and all the things I think we now call "sovereignty". But there was lots of theory and precious few solid, documented studies. There were also hundreds of propaganda pieces and Google/Microsoft/Amazon shill pieces sowing disinfo.

Surely that's changed now and the economic realities are clearer?

Did you find differences in performance or reliability of the S3 and Kubernetes services already?

The reliability of OVH has so far not been a problem. Performance is mostly a factor of the VM instances that you use, AWS has much more instances to choose from, but obviously at a much higher cost, so performance per dollar is obviously better with OVH.

That's already answered in OP's article.

k8s:

> Good stability.

S3:

> AWS S3 is the premier cloud-based object storage service. It may have higher availability, but it’s three times more expensive than OVH’s S3 storage.

Any concrete plans to offer Serverless also in Europe?

We are already available in Europe; it is just our infrastructure that is currently more us-centric in that sense.

Eventually as the demand grows we will of course deploy in EU region to improve the availability.

If it is more of a sovereign aspect; we are deployable on kubernetes;one can already use our installer on OVHCloud and deploy on a European region.

Why OVH and not for instance Hetzner?

One obvious benefit of OVH compared to Hetzner is the S3 storage, the Kubernetes framework and a number of other services provided by OVH, don't think Hetzner would provide those services.

Hetzner has S3-like storage, https://www.hetzner.com/storage/object-storage/.

Hetzner's Object Storage does not seem API complete and it was released only recently.

Long time Hetzner user, but I would not trust new system (S3 storage) in this case for things beyond playground and tests until it has at least couple of years in production - Hetzner just made it public in ~ Nov 2024.

How large was the total data transfer? It doesn't seem to have been that big.

You'd be surprised :) We had to migrate a few thousand projects and their feature data and users' custom images. It was substantial. (eng@Hopsworks)

So I guess this opens the question of which part really covers MLOps; I would love to see those but some strike me heavily as being part of the model development and training. I somewhat, in my simple mind always got stuck on the Ops in a “how to keep the system rolling” kind of way.

Life finds a way

In a nutshell; no infra means infra but not managed by yourself; so you would focus on all the different ML pipeline in the journey (feature, training, inference) to create a real operationalised ML system.

Set of general principles and practices in an organisation.

I’ve participated in the first two lectures: it’s perhaps the most straight forward approach that I had to get an end to end ML service running.

While this article is on offline testing with pytest we are also preparing a second one on continuous integration tests.

Well, all good and nice, but we are so far behind on data infrastructure, development, and investment that I do not know how realistic it is to set those targets when the basics are not here.

ok.