Understanding AI Optimizer for IBM Z and IBM LinuxONE installation

Installation architecture

AI Optimizer for Z and LinuxONE runs as a containerized appliance on a dedicated IBM Secure Service Container (SSC) LPAR. The installation process involves two main phases:

Phase 1: LPAR preparation and image installation

You configure the SSC LPAR in installer mode and deploy the AI Optimizer for Z and LinuxONE appliance image. This phase installs the base operating system, runtime environment, and all appliance components onto the LPAR.

Phase 2: Appliance configuration and component initialization

After the image is installed, you use the AI Optimizer for Z and LinuxONE Appliance Manager UI to configure the container registry, initialize all system components, and verify that the appliance is ready for use. This phase is common to all installation methods.

The SSC LPAR provides an isolated, secure environment that protects the appliance and its workloads. The LPAR must be configured with sufficient resources, including CPU, memory, storage, and attached IBM Spyre Accelerator accelerator cards for AI inference workloads.

Installation methods

AI Optimizer for Z and LinuxONE supports three installation methods, each suited to different deployment scenarios and automation requirements:

IBM Hardware Management Console (HMC) installation

A manual, interactive installation method using the HMC interface. This method is ideal for initial deployments, proof-of-concept installations, or environments where you want direct control over each installation step. The HMC provides a web-based installer interface that guides you through the process.

Appliance Control Center for IBM Z and LinuxONE (ACC) UI installation

A web-based installation method using the ACC user interface. This method simplifies the installation process by providing a centralized management interface for multiple systems. It is suitable for environments where ACC is already deployed and you want to manage AI Optimizer for Z and LinuxONE installations alongside other appliances.

ACC Ansible Playbooks installation

An automated installation method using Ansible playbooks that interact with the ACC API. This method is ideal for large-scale deployments, standardized installations across multiple systems, or integration with existing automation workflows. The playbooks enable repeatable, consistent installations with minimal manual intervention. For sample playbooks, see the AI Optimizer for Z and LinuxONE Ansible playbooks repository.

Regardless of which installation method you choose, the final configuration steps are identical. After the appliance image is installed, you use the Appliance Manager UI to configure the container registry and initialize all system components.

Container registry options

During configuration, you must set up a container registry to provide access to AI Optimizer for Z and LinuxONE component images and LLM models. The system supports two registry types:

IBM Container Registry

The default registry option for environments with internet connectivity. This registry is hosted by IBM and provides direct access to all AI Optimizer for Z and LinuxONE images. You authenticate using your IBM Entitlement Key, which you obtain from the IBM Container Library. This option simplifies installation and ensures you always have access to the latest images.

Custom registry

A private container registry that you host and manage in your own environment. This option is required for air-gapped or restricted network environments where direct internet access is not available or not permitted. Before installation, you must mirror all required AI Optimizer for Z and LinuxONE images from the IBM Container Registry to your custom registry. The custom registry supports authentication using either username and password credentials or access tokens, and can be configured to use proxy servers if needed.

The registry configuration you choose affects how the system accesses and downloads component images during initialization and LLM deployment. Ensure that your selected registry is accessible from the SSC LPAR and that you have the required authentication credentials before starting the installation.

End-to-end workflow

The complete installation and deployment workflow consists of the following stages:

1. Planning and preparation: Allocate hardware resources, obtain required credentials and materials, and plan your registry configuration. See Preparing for AI Optimizer for IBM Z and IBM LinuxONE installation for detailed planning guidance.
2. Image mirroring (if using custom registry): Mirror all required images from the IBM Container Registry to your custom registry. See Mirroring container images to a custom registry for the mirroring procedure.
3. LPAR configuration and image installation: Configure the SSC LPAR and install the appliance image using your chosen installation method. See Installing and configuring AI Optimizer for IBM Z and IBM LinuxONE for the installation procedure.
4. Container registry configuration: Configure access to your container registry (either IBM Container Registry or custom registry) through the Appliance Manager UI.
5. Component initialization: Initialize all system components and verify that they are running correctly. Monitor the status indicators until all components show green status.
6. LLM deployment: Deploy one or more LLM models to begin using AI Optimizer for Z and LinuxONE for AI inference. See Deploying or undeploying LLMs in AI Optimizer for IBM Z and IBM LinuxONE for the deployment procedure.
7. Verification and ongoing management: Verify that the system is operational, monitor system health, and perform ongoing management tasks as needed.

Each stage builds on the previous one, and you must complete them in sequence. If you encounter issues during any stage, you can access system logs and troubleshooting resources to identify and resolve problems. See Viewing and downloading system logs in AI Optimizer for IBM Z and IBM LinuxONE for information about accessing logs and Troubleshooting problems in AI Optimizer for IBM Z and IBM LinuxONE for troubleshooting guidance.