Installing the Custom Metrics Autoscaler
By default, the Red Hat® OpenShift® Horizontal Pod Autoscaler (HPA) uses resource-based automatic scaling to increase or decrease the number of pods in response to CPU or memory consumption based on the utilization threshold.
If you plan to install services with a dependency on Inference foundation models, you can install Red Hat OpenShift Custom Metrics Autoscaler to enable event-driven scaling on GPU. Event-driven scaling enables the cluster to automatically scale model replicas across existing GPU nodes in response to inferencing requests.
- Installation phase
-
Setting up a client workstation
Setting up a cluster
Collecting required information
Preparing to run installs in a restricted network
Preparing to run installs from a private container registry
Preparing the cluster for IBM Software Hub
Preparing to install an instance of IBM Software Hub
Installing an instance of IBM Software Hub
Setting up the control plane
Installing solutions and services
- Who needs to complete this task?
- Cluster administrator A cluster administrator must complete this task.
- When do you need to complete this task?
-
One-time setup This task is optional but recommended if you plan to install one or more services that use Inference foundation models to start and host foundation models:
- IBM Knowledge Catalog Premium *
- IBM Knowledge Catalog Standard *
- Watson Speech services *
- watsonx.ai™
- watsonx Assistant *
- Watsonx BI
- watsonx Code Assistant™
- watsonx Code Assistant for Red Hat Ansible® Lightspeed
- watsonx Code Assistant for Z Agentic
- watsonx Code Assistant for Z Understand
- watsonx.data™ Premium
- watsonx.data integration *
- watsonx.data intelligence *
- watsonx™ Orchestrate *
About this task
The Red Hat OpenShift Custom Metrics Autoscaler automatically installs the Kubernetes Event-driven Autoscaler (KEDA).
- If there are no new inferencing requests, the cluster scales down the number of model replicas to free up GPU.
- If the number of inferencing requests increases, the cluster scales up the number of model replicas on available GPU to maintain latency.
- Enable event-driven automatic scaling on GPUs for the instance.
- Configure event-driven scaling for models that are running in the instance.