Optimized Inferencing and Integration with AI on IBM zSystems: Introduction, Methodology, and Use Cases

This IBM Redpaper publication introduces and explains AI technologies and hardware optimizations, and demonstrates how to leverage certain capabilities and components to enable AI solutions in business-critical use cases, such as fraud detection and credit risk scoring, on the platform. Real-time inferencing with AI models, a capability that is critical to certain industries and use cases, now can be implemented with optimized performance thanks to innovations like IBM zSystems Integrated Accelerator for AI embedded in the Telum chip within IBM z16TM. This publication describes and demonstrates the implementation and integration of the two end-to-end solutions (fraud detection and credit risk), from developing and training the AI models to deploying the models in an IBM z/OS® V2R5 environment on IBM z16 hardware, and integrating AI functions into an application, for example an IBM z/OS Customer Information Control System (IBM CICS®) application. We describe performance optimization recommendations and considerations when leveraging AI technology on the IBM zSystems platform, including optimizations for micro-batching in IBM Watson® Machine Learning for z/OS. The benefits that are derived from the solutions also are described in detail, including how the open-source AI framework portability of the IBM zSystems platform enables model development and training to be done anywhere, including on IBM zSystems, and enables easy integration to deploy on IBM zSystems for optimal inferencing. Thus, allowing enterprises to uncover insights at the transaction-level while taking advantage of the speed, depth, and securability of the platform.

This publication is intended for technical specialists, site reliability engineers, architects, system programmers, and systems engineers. Technologies that are covered include TensorFlow Serving, WMLz, IBM Cloud Pak® for Data (CP4D), IBM z/OS Container Extensions (zCX), IBM CICS, Open Neural Network Exchange (ONNX), and IBM Deep Learning Compiler (zDLC).