Adaptive Kafka Resource Allocation for Real-Time Monitoring in the Manufacturing Sector

The deployment of Industry 4.0 technologies in the manufacturing sector creates an immense demand for scalable and efficient real-time data processing. This paper presents an adaptive resource allocation framework to address these challenges. We introduce a methodology for creating and managing adaptive Kafka clusters, encompassing broker, partition, and consumer scaling. The framework's core for dynamic resource provisioning is an algorithm named Bromin, supplemented by KEDA for consumer autoscaling. We first establish the importance of empirical, hardware-specific parameterization, showing that our tuned configuration consistently meets throughput target where configurations based on industry-standard estimates fail. We then evaluate the fully adaptive cluster under various loads. While throughput remained high, the study revealed critical limitations in consumer scaling under extreme load, leading to a complete halt in message processing. Finally, we demonstrate the framework's practical application in an end-to-end real-time monitoring and alerting system.