Cloud Platform as a Service

As most already know, Red Hat OpenShift AI provides much of the underlying capabilities found in IBM's watsonx.ai. If your needs are best met by OpenShift AI and you want to quickly create a Red Hat OpenShift AI deployment in your IBM Cloud account, this blog is for you.

IBM Cloud has the concept of Deployable Architectures which describes them exactly, an architecture you can deploy. In our case we want a ROKS cluster that has the OpenShift AI stack of operators installed and ready for use. We can use a deployable architecture to quickly produce this environment for us. The good news is that you can create a Private Catalog where you can install a Deployable Architecture. We will leave that exercise for another blog post, but we can go over how to execute our Deployable Architecture catalog entry once you get it into your catalog.

First, you can find this Deployable Architecture here. You will notice it is mostly standard Terraform with a few extra files. You are free to execute that Terraform in your own environment, but let's look at how you would execute this Deployable Architecture.

In your catalog, we first have to find our Deployable Architecture tile. You can narrow them down by a filter:

You should now be able to find the tile we are looking for:

At the bottom of the page, click the "Review deployment options" blue button.

We will be deploying using Terraform and the IBM Schematics service, so click the "Deploy with IBM Cloud Schematics" deployment option selection. Make sure you have the necessary permissions to execute this DA.

Configure your Schematics workspace to your liking.

And now provide the input parameter values.

1. ibmcloud_api_key - your API key
2. cluster-name - provide a name for your new cluster
3. region - the region to create the cluster in
4. number-gpu-nodes - how many GPU worker nodes do you want
5. create-cluster - create a new cluster?
6. ocp-version - specify the OpenShift version
7. machine-type - worker node flavor
8. cos-instance - an existing Cloud Object Storage instance in your account

Don't forget that this cluster costs real money. Please understand the cost of the cluster you intend to create before executing this DA. GPU worker nodes cost more than standard workers, keep that in mind. The goal of this DA is to help you easily create a simple cluster, but once it is created, you are responsible for it and its cost. The DA cost summary will show you an example monthly cost for a 2 node gx3.16x80.l4 cluster. 

Note: You may have a GPU based cluster already created that you want to use. Simply set the cluster-name input parameter to the existing cluster name, specify the region that the existing cluster resides in, set the number-gpu-nodes parameter to the number you expect the DA to find in the cluster, and set the create-cluster parameter to false. The DA will install the OpenShift AI software into your existing cluster.

Now that you have specified your input parameters, check the license checkbox and click "Deploy." You will see a schematics workspace get created and can watch the deployment via the log. One thing to note. As we all know, pods and when they complete in a cluster is not a consistent experience. There are times when timing is off and something in the Terraform may fail due to things not yet complete. It does happen from time to time. Check the Schematics log for details, but try to simply apply the Schematics plan again before assuming you have a problem.

Once complete, you will have a ROKS cluster in your cloud account with the full set of OpenShift AI operators installed ready for you to explore what OpenShift AI can do. Try out the OpenShift AI learning paths to learn more. Have fun.

UPDATE: You will see that now when you click on a Deployable Architecture tile your deployment options do not include IBM Schematics as an option. You are encouraged to create a project and then run the DA terraform in the project. Nothing has really changed but Projects are where all the new DA and DA stacks will be realized.