Apache Flink Migration for Real-Time Data Streaming Services: Methodology and Evaluation

Abstract

The continuous growth in volume, velocity, and diversity of data generated by vehicles and automotive production systems has exposed the limitations of traditional batch processing architectures and microservice-based approaches. In response, stream processing frameworks such as Apache Flink have emerged as promising solutions for building scalable, low-latency, and resource-efficient data pipelines. This dissertation presents the controlled migration of an existing automotive data processing service within BMW's connected vehicle ecosystem, built on the Quarkus framework, to an Apache Flink-based stream processing architecture. The work encompasses requirements elicita- tion of the existing system, incremental architectural design of the Flink-based solution, functional validation through module testing, and performance benchmarking comparing the legacy system with the migrated implementation. The analysis of existing literature reveals a clear research gap: while Apache Flink is increasingly adopted in large-scale automotive data pipelines, no studies systematically document the migration of an existing automotive service to Apache Flink with quantitative benchmarks and reusable migration guidelines. This work addresses this gap by contributing validated per- formance evidence, cost analysis, and a set of migration guidelines applicable to automotive and other data-intensive industrial systems.