The stream processing of data and the usage of modern unsupervised ML algorithms allows us to detect, cluster and find the root-cause of cascading failures in big interconnected software and hardware landscapes in real-time. We display them in dashboards and act on them via automated maintenance tasks. A chat replaces a traditional alert mailing list. All the used tools are Open Source, the algorithms and the architecture can in principal be adapted to any environment. We anticipate that AI-based monitoring solutions will serve the industry.
Target Audience: Decision Makers, Data Scientists, DevOps, Project Leaders, Founders, AI Enthusiasts, Students
Prerequisites: Basic understanding of machine learning and IT infrastructure
The existing computing capacities, the developed algorithms and the possibility to store and process a large amount of data enabled the usage of Machine Learning and self-learning algorithms. Nowadays streaming technologies enable to analyze a large amount of data in real-time. Furthermore, today's software and hardware architectures are as distributed as they are interconnected. Such complex systems need a sophisticated monitoring solution to ensure a smooth operation and cost efficiency. We show how one can use machine learning to find, understand and quickly act on cascading failures in big interconnected software and machine landscapes. We present a custom ML solution and an open source tool-chain which addresses this challenge. This AI-powered monitoring can compliment traditional monitoring solutions. In segmenting and grouping problems, it has proven to be superior and fills the gaps of traditional strict rule-based monitoring solutions. To ensure a fast incident response time, automated counteractions are triggered and a chat is used to report alerts. We demonstrate what is possible with modern streaming technologies and machine learning algorithms on a large-scale productive system. It is anticipated that this and more sophisticated future AI-based monitoring solutions, will serve the industry worldwide to ensure a more efficient operation.
Diese Session stellt das Predictive-Analytics-Projekt zur automatisierten Vorhersage der Bestandsveränderung vor. Im Rahmen dieses Projektes wurden verschiedene Methoden der Data Science einander gegenübergestellt und bewertet. Dabei spielt insbesondere der Transfer dieser Algorithmen in die Unternehmenspraxis eine zentrale Rolle.
Zielpublikum: Data Scientist, Anwender, Projektleitung und Strategie
Voraussetzungen: Grundkenntnisse Data Science, Projekterfahrung wünschenswert