1. Google will do this anyway, because it's not going to run on AWS or azure, so why not keep some skin in the game?
2. Isn't k8s a play to reinvent this space?
2. Isn't k8s a play to reinvent this space?
Amazon and MS have vast support networks because enterprise wants to feel like they have some control and feedback on their systems.
Also, unless you can innovate in the space (or go really cheap), why would I ever choose a smaller player over a larger one?
This is another example of the problem these rich companies have where cash is not a problem but identity and direction is.
But, also in think Google does have some advantages. Out of the box, their NLP algorithms seem much better than AWS’s offering. They also offer custom search as an api. I haven’t used it, but bigquery seems very good. I do agree these are probably not enough to have an enterprise shift. Maybe if they specialize on a couple industries that won’t choose amazon (fedex, ups, target, Walmart, streaming/media). It does seem like google has made strong inroads in finance as well.
Azure is doing quite well leveraging Microsoft's existing enterprise relationships. I'm not sure how much success folks are having "lifting and shifting" their Microsoft infrastructure into the cloud, some, maybe?
However, everyone that I've heard of that understands cloud native type architectures that has attempted to deploy into Azure is looking to leave.
Extremely unsurprising, but I'd be interested to read more.
However, that correlates pretty closely with my own personal experience. I've worked with a fair bit of the Azure products and it feels to me very much like a work in progress.
The identity management stuff has been a moving target since day one. That is, how to authorize against the service itself. This is apparent in many of the Python libraries where this is documented.
The Python libraries that I've seen look like they were produced by interns. I'm performance with ADL is pretty rough.
My main goal with Azure was to deploy Kubernetes clusters there. The list of problems is very long and has been documented ad nauseum in many Medium articles. In the 14 or so months I've been working with Azure, it hasn't improved as quickly as I'm accustomed to with AWS and GCP.
A year ago, I shifted the majority of my production infrastructure from AWS to GCP. By comparison to Azure, I was pleasantly surprised with the quality of the products from GCP. My company uses GSuite which made identity management and RBAC a breeze. The GKE product is way, way better than anything else, and there are a handful of things that I like better than AWS.
I wouldn't encourage anyone to shift from AWS unless they were planning a large Kubernetes implementation. I would encourage folks to try GCP and especially GKE to get a sense of the differences.
There's nothing about Azure that I've discovered yet that would cause me to recommend it to anyone. It's possible that if a company had a heavier investment in MS products that it might be worthwhile, but that's not the case for mine. We only ever use MSSQL occasionally for development purposes and I've found that I can spin up an instance of that quite nicely in k8s with the Microsoft authored helm chart.
Also once you start getting into k8s you will find so much more information in the community with specific instructions for GKE; its by far the best environment to learn in for that reason.
Google falls down in so many other areas though that it probably doesn't matter how good GKE is. The biggest annoyance to me is their Postgres offering still has no point in time recovery.
edit to add: One other big advantage I forgot to mention is that launching a GKE cluster takes only a couple of minutes. I have a Terraform configuration that will launch a GKE cluster and run my deployments and my app is up and running after less than 5 minutes. This makes it much more attractive to do things like spin up a new cluster for each pull request.
I will agree that EKS startup is slow, but given the way it needs to run, I don’t think many customers will want to run more than a few clusters per VPC.