Guests: Mohit Dadu (LinkedIn | John Hart (LinkedIn)
Company: Broadcom (Twitter)
In this episode of TFiR Mainframe Matters, recorded at the Open Mainframe Summit in Philadelphia, Swapnil Bhartiya sits down with Broadcom representatives Mohit Dadu, Senior Manager of Data Science & Engineering, and John Hart, Data Scientist to discuss artificial intelligence/machine learning (AI/ML) for monitoring in mainframe.
Key highlights from this video interview:
- Dadu discusses how the increase in transactions has meant that more load is being put on mainframes. He explains how they are looking to ML to help streamline the processes and cut down the number of alerts their product generates so that people can get faster root cause analysis and find the problem and resource it quicker.
- From a data scientist perspective, the mainframe has several challenges, including that the CPU memory is extremely expensive compared to distributed, so the models tend to be smaller. Hart discusses the IBM Telum chip and how this is changing the kind of models that can be used.
- When Broadcom started its AI Ops journey, the biggest challenge operators faced was around static thresholds, such as alerts when a CPU goes above 90%. The company first looked into removing these rules and automatically learn based on historical data with an anomaly detection model. Then, they looked at clustering together similar alerts.
- For companies in the mainframe space who want to leverage ML technologies to improve their monitoring or security, Dadu recommends a phased manner. Cut down the number of rules and instead use anomaly detection. Then, augment SME knowledge and where there are gaps, fill it with machine learning. From there, take advantage of clustering.
- The barrier of entry is still around perception. ML is just using mathematical tools to simplify our lives. For example, with Grammarly, we do not have to learn all the grammar rules – it takes care of it.
- Dadu feels that there is no difference between mainframe and distributed: it is just another platform where you can run Python or docker containers. Yet, the mainframe is often viewed as a legacy platform even though the modernization efforts that are happening with the cloud are also happening with the mainframe.
- Research has shown that the mainframe is on a par with a distributed environment in terms of digital transformation. Each year, they are adopting the latest technologies, such as IBM’s newest chip Telum.
The summary of the show is written by Emily Nicholls.