Setting up a dedicated node for models running on CPU

To improve the performance of models that run on, you can dedicate a worker node to the pods on which models with runtime are deployed. The node is then used exclusively by these pods.

Required roles

A cluster administrator must complete this task.

Procedure

Some of the following commands use environment variables so that you can usually run the commands exactly as written. For information about sourcing environment variables, see Setting up installation environment variables.

Complete these steps:

Log in to the cluster as Red Hat OpenShift Container Platform cluster administrator with the username kubeadmin.
Open the namespace where IBM Knowledge Catalog is installed:
```
oc project ${PROJECT_CPD_INST_OPERANDS}
```
List all nodes in the namespace where IBM Knowledge Catalog is installed:
```
oc list nodes
```
Select the node that you want to use for the model deployment from the list that was returned in the previous step.
Taint the node by running the following command. Replace worker_name with the name of the selected node:
```
oc adm taint nodes <worker_name> cpu_llm_model=true:NoExecute
```
Label the worker node for affinity by running the following command. Replace worker_name with the name of the selected node:
```
oc label node <worker_name> cpu_llm_model=
```
Check whether the changes are applied to the worker node:
```
oc describe node <worker_name>
```
The Labels section of the returned information should have a cpu_llm_model= entry. The Taints section, should have a cpu_llm_model=true:NoExecute entry.