z/OS AI Framework components
The AI Framework for IBM® z/OS® consists of 5 major component areas that, together, enable AI-enhanced capabilities within the z/OS system.
Figure 1 shows the major components that form the AI Framework for IBM z/OS.
The following component descriptions correspond to the numbered components shown in Figure 1:
1 — Data collection
Data collection provides a common way to collect IT data for use in AI model training.
- Collects and parses the raw IT data.
- Streams the data to the data store.
AI Framework for IBM z/OS uses IBM Z® Common Data Provider (ZCDP) as the data collection engine.
- Implemented using VSAM data sets.
- Accessed by the model training pipeline and, optionally, by the deployed model.
- Can be used both for training and for the deployed model to store its own data.
2 — AI model server
- Manages model training, versioning, deploying, and monitoring.
- Supports failover for high availability.
- Accessed by the system via REST APIs.
AI Framework for IBM z/OS uses Machine Learning for IBM z/OS Core Edition (MLz Core) as the AI model server.
3 — AI Base Component
- Access is via traditional z/OS assembler services (macros).
- Handles connection to REST APIs.
4 — User interface (z/OSMF)
- A z/OSMF workflow, known as the AI Framework for IBM z/OS Configuration Workflow, guides you through the installation process with detailed configuration steps for each of the framework components.
- The AI Control Interface for IBM z/OS, a new task on the z/OSMF desktop, provides AI model management. You use the z/OS AI Interface to initiate training of a model for a use case and to enable or disable the AI mode for that use case or place it into simulation mode.
5 — AI providers and use cases
- Define data collection.
- Create the model training pipeline.
- Call the z/OS AI Base component.
- Extend the user interface.
Workload management (WLM) is the first provider of an AI use case, AI-powered WLM batch initiator management, that uses the z/OS AI Framework. The use case proactively starts WLM-managed initiators based on predictive insights on upcoming batch workloads.
The framework is designed to be expandable to include additional providers and use cases.