A machine learning based capacity management system for mainframe resources

Abstract

The goal of this study is to design a capacity planning tool for resource consumption of application servers which are running on mainframes, also known as Z systems, by using machine learning algorithms. This tool is aimed to ensure adequate resources are available in order to meet current and future workload demands. The desired system is intended to have capability to determine and then forecast how much additional capacity will be needed based on increasing demands. In this study, IBM Cloud Pak for Data as a Service is used to create capacity planning model by using data analysis, data engineering, data governance and Artificial Intelligence modeling services which are provided by the platform. The data is prepared outside of the platform and imported to the platform in order to perform analysis and refinement. After the data refinement step is completed, machine learning models are trained by using several algorithms. Then, functional tests are performed in order to check accuracy and performance of the models by using the test interface of the platform. Results of these tests, comments and further research opportunities are also provided. It is observed that the designed capacity planning tool is capable of making consistent predictions with acceptable error rates.