In February of this year, IBM announced the release of IBM Machine Learning for z/OS. As with most exciting new Z technologies, such as z/OS Connect Enterprise Edition, IMS is officially supported with the initial release. I want to take some time in this blog post to discuss how machine learning adds value to your existing IMS assets (both data and transactions), and to share some new programming paradigms that are now possible, such as Hybrid Transactional/Analytical Processing (HTAP) applications.
IBM Machine Learning on z/OS and IMS
Machine learning has been around for a while, so I won’t go into too much detail on its capabilities. At a high level, you can think of machine learning as a way to build scoring models based on your data. For example, if you want to identify the likelihood of customer churn, you can build a model based on historic data to score each customer on whether or not they are satisfied with your services.
With most other machine learning solutions, you typically have to move your data to where the analytics engine is to build your models. By providing a solution on z/OS, IBM makes it possible for mainframe users to keep their enterprise data on the mainframe while building out their scoring models. This allows your potentially sensitive data to stay on a secured environment aka the mainframe. It also has the added benefit of avoiding potentially costly extract, transform and load (ETL) operations related to moving data.
So how does this work with IMS? We can look at this from two perspectives, depending on whether you’re a user of an IMS database or an IMS transaction.
Building a scoring model with IMS data
In order to build a model, IBM Machine Learning on z/OS needs to be able to access your data. This access is provided through IBM Mainframe Data Services for Apache Spark for z/OS (MDSS). You can sign up for a free 30-day trial of MDSS here. MDSS reads the flat files that your IMS data is stored on and puts it directly into a Spark DataFrame. After your data is in a data frame, you can use one of the supported languages, such as SCALA, to both curate your data and build your scoring model. This figure shows the IBM Machine Learning on z/OS architecture and how IMS data gets fed in.
Invoking a scoring model from an IMS transaction
After you you build a scoring model, the next step is to leverage it in an IMS transaction. IBM Machine Learning on z/OS makes your scoring models available as RESTful services. The great thing about the RESTful service model is that it’s fairly language-agnostic; many programming languages provide ways for you to invoke them. Specifically for IMS, this means that you can invoke your prediction model from a new transaction written in Java or even from a modified COBOL transaction.
The value of HTAP IMS applications
You might be thinking, “So, great, I can invoke my prediction models from my IMS transaction, but what does that buy me?” Quite a lot actually. HTAP allows for “in business real time” decision making. This means that your point of sale transactions can do a lot more analytical processing.
A great example that impacts nearly all of us is fraud detection. Typically, banks do nightly dumps of their data to search for fraudulent activity. This can mean a lengthy turnaround, up to 24 hours, before fraudulent activity is detected. What if you can detect this activity at the point of sale, when the transaction is being processed, prior to payment processing? Early detection can lead to a decrease in the amount of money the bank would pay out to fraud victims.
Testing the waters
If you’re fairly new to the concept of machine learning, IBM provides a sandbox environment through the IBM Data Science Experience where you can take a stab at curating your own data sample sets and building your own scoring models. This is a great way to get your feet wet and explore what potential insights you might be able to get out of your data.
As always comments are always appreciated! If you have other scenarios, even hypothetical ones, for how machine learning and HTAP might benefit enterprise data, feel free to let us know.