0

Since we are working on ML engineering tasks we do have requirements of gpu instances. We do have couple of them on-premise and we need to organize our pipeline creating a kubernetes cluster.

Since these gpu instances remain on-premise our use case is to a. create a kubernetes master on google cloud and b. connect this master to our private network.

We would like to know if this is feasible and what kind of google services we need in order to start with. In addition, we want to use our private network as the main network and not route all our traffic through VPC.

Any ideas ? Thanks you in advance.

Nikolaos Epitropakis
  • 1

1 Answers1

0

If you're spinning up a cluster using GCE instances (as opposed to have a managed GKE cluster), then as long as your on-prem network can reach the GCP VPC network where your instances are should make it feasible.

For this purpose, you can use a VPN connection with Cloud VPN.

As for the traffic, if all your instances are on-prem, you'll only have to worry about the connections flowing between the control plane on the VPC to your private network nodes.

For routing, it'll depend on how you set up your VPN and what kind of control you need to have. Currently Cloud VPN supports dynamic routing with BGP or static routes.

So answering your questions: It is feasible and these are the Google services that you can use to achieve that.

yyyyahir
  • 265
  • 1
  • 6