Addressing IT Capacity Management Concerns Using Machine Learning Techniques

Abstract

The digitization of various industries is emerging from and being supported by the Information Technology (IT) industry. However, bringing about the practical implementation of the fourth industrial revolution will require different fields of IT to undergo their own transformations. One of the research fields that draws a lot of attention nowadays is machine learning and its application in different areas. Therefore, in this paper, we present an analysis on the applicability of various machine learning techniques to address different problems in the field of capacity management for Commercial-off-the-shelf enterprise applications. Our investigation of the selected machine learning techniques is based on real monitoring data from over 18,000 SAP applications and database instances that are hosted on more than 16,000 different physical servers. These data are used to train various performance models, such as support vector machines with different kernels, random forests, and AdaBoost, for standard business functions. Boosted trees achieve sufficient accuracy to predict mean response times for ten frequently used transactions. To evaluate the suggested models, we applied them successfully to address different concerns in the context of capacity management. The evaluation includes multiple scenarios in the fields of server sizing, load testing, and server consolidation, with the objective to identify cost-effective designs. Based on the same monitoring data, we also present an anomaly detection scenario. In this scenario, we aim to demonstrate the use of machine learning techniques on historical data to detect possible performance anomalies for a suggested design or even predict possible anomalies in future scenarios. Results strongly emphasize the need to integrate monitoring data from standardized business applications to allow for novel and cost-effective capacity management service offerings.