Mainframes are complex. CICS, IMS, and Db2 are each complex in their own ways. So, how can you get the most out of your mainframe? One way is to have lots of very bright and well-trained staff monitoring and controlling what’s happening on your mainframes and network. This, of course, can be costly, and has the added problem that many of your really experienced staff are getting close to retirement age. When they retire, they will take all of those years of experience with them. The alternative is to investigate running some kind of artificial intelligence (AI) on your mainframe. And that’s where IBM Watson Machine Learning for z/OS (WMLz) comes in.
To be clear, WMLz can’t run your mainframe for you, but it is taking baby steps toward making some applications (e.g. Db2) run more efficiently. In effect it is working like a DBA on steroids!
What Is Watson?
Watson first came to the public attention in 2011 by winning a game of Jeopardy. The hardware used was ninety IBM Power 750 servers with 16 terabytes of RAM. It used IBM’s DeepQA software and the Apache Unstructured Information Management Architecture (UIMA) framework implementation. The information that Watson used in the game came from encyclopedias, dictionaries, thesauruses, newswire articles, literary works and other lists of information. Since 2011, Watson has been hugely enhanced and used in a variety of different industries. There’s talk of it being a better diagnostician than a doctor or a psychiatrist were you to discuss your problems with it. It’s meant to be able to work in a call center, where it would have to be able to successfully deal with a variety of customer and potential customer enquiries.
Watson and Db2
So, where does Watson fit in? Let’s focus on Db2 for the moment. The IBM Db2 AI for z/OS (Db2ZAI) can optimize a Db2 for z/OS engine to determine the best-performing query access paths, based on the workload characteristics. The optimizer consists of Relational Data Services (RDS) components that govern query transformation, access path selection, run time and parallelism for every SQL statement used. The access path for a SQL statement specifies how Db2 accesses the data that the query specifies. It determines the indexes and tables that are accessed, the access methods that are used, and the order in which objects are accessed. Db2ZAI collects data from the optimizer and the query execution history, finds patterns from this data and learns the optimal access paths for queries entering Db2 for z/OS. That means data access is as quick as it can possibly be. And because each site has a unique workload, this optimization is specific for that particular site. Should workloads change, the optimization will change, too. It will learn how to work the best it can.
Once trained, the model is ready to be deployed into production, providing insights to the optimizer’s access path selection. These insights are in addition to what the optimizer uses today in the selection of the best query path. The information is unique to each specific environment, and currently unknown to the traditional query optimizer.
Learning how things work in a particular computing environment is called machine learning, and machine learning is a subset of AI.
It was back in 2017 that IBM announced that it would add the machine learning technology from Watson to z/OS to provide smarter and faster analytics of transaction data. The gap that it was trying to fill was the one between mainframes and big data analytics. Watson’s job was to sort and transform mainframe data in order to perform complex data analytics—something most sites weren’t doing. According to IBM: “IBM Machine Learning for z/OS empowers IBM Z systems customers to create, deploy and manage high-quality self-learning behavioural models to extract hidden value from enterprise data securely, in place, and real time.”
Simplifying Production Implementation of AI Models
In 2019, WMLz, currently at Version 2.1, is said to simplify the production implementation of AI models. Users can develop models where they want. And, users can readily deploy within their transaction applications for real-time insight.
WMLz features include:
• Flexible model development. Data science teams have the flexibility to build, train and evaluate models using their Integrated Development Environment (IDE) of choice or use the WMLz extensive model building features.
• Improved productivity. WMLz offers several model building modes including notebooks, visual builders, wizards and enhanced intelligence. The product can automatically normalize, handle missing values, and generate data features.
• Enterprise-ready AI model deployment. WMLz offers several scoring approaches including RESTful APIs and Java and CICS integration, optimized for the highest security and performance levels on IBM Z.
• Enhanced model accuracy. Users can schedule continuous reevaluations of new data to monitor model accuracy over time and be alerted when performance deteriorates.
• Production-ready machine learning. With model versioning, auditing and monitoring as well as high availability, high performance, low latency and machine learning model automation (ML as-a-service).
• Quick-start solution templates. WMLz comes with templates for common business requirements solution templates demonstrate how machine learning can run alongside your application infrastructure to add value to key business areas including fraud detection, loan approval and IT operational analytics (ITOA).
At the moment, WMLz can help sites optimize their transactions, but it isn’t restricted to just mainframes. WMLz offers a hybrid cloud approach to model development and model deployment lifecycle management and collaboration that is designed to help organizations innovate and transform on an enterprise scale. Db2ZAI is a separate product that can optimize data access in Db2. I would hope that we will see something similar for CICS and IMS in the near future. And, in the not too distant future, we will, perhaps see more features on the mainframe optimized and controlled by Watson.