Provision a Tanzu Kubernetes Grid Cluster from the VMware Aria Automation Self-Service Catalog for Private AI Ready Infrastructure for VMware Cloud Foundation

As a DevOps engineer, use the Automation Service Broker catalog to provision a GPU-enabled Tanzu Kubernetes Grid cluster with the NVIDIA GPU Operator configured and licensed, and ready to run GPU-enabled container workloads.

Procedure

1. Log in to VMware Aria Automation at https://<aria_automation_cluster_fqdn>/csp/gateway/portal.
2. On the main navigation bar, click Services.
3. On the My Services page, click Service Broker.
4. On the Consume tab, on the navigation bar, click Catalog.
5. In the AI Kubernetes Cluster card, click Request .
6. Configure the following settings and click Submit.

   Setting	Value
   Version	Version of the catalog item
   Project	Project where you want it to be deployed and the name for the deployment
   Deployment Name	Name for the resulting deployment
   Control plane > Node count	Number of control place nodes
   Control plane > VM Class	VM class, based on CPU and memory requirements, for the control plane nodes
   Workers > Node count	Number of worker nodes
   Workers > VM Class	GPU-enabled VM class for the worker nodes
   NVIDIA AI enterprise API key	The API key for access to the NVIDIA NGC registry. The API key is required to download the Helm charts of NVIDIA GPU Operator.

7. Monitor the deployment process.
   1. On the Consume tab, click Deployments > Deployments.
   2. Click the name of deployment and then click the History tab.
8. After the deployment is completed, review the details to access the AI-ready TKG cluster by using kubectl.

   Figure 1. Example of Access Details for an AI -Ready TKG Cluster

   
   
   

9. Verify the NVIDIA GPU Operator deployment by running the following kubectl command on the Supervisor.

   kubectl get pods -n gpu-operator

   Figure 2. Example of Successful NVIDIA GPU Operator deployment

   
   
   

10. Verify the NVIDIA license by running the following command.
    You run a Bash shell into the nvidia-driver-daemonset pod where you can run the nvidia-smi command with the -q argument to check the license status.

    kubectl exec nvidia-driver-daemonset-x6jbv -n gpu-operator -it -- /bin/bash

    Figure 3. Example of a Licensed AI Kubernetes Cluster