Unveiling the Timing Behavior of the Data Distribution Service: Modeling, Analysis and Optimization of Real-Time DDS-Based Distributed Systems

The successful design and operation of real-time systems heavily rely on rigorous analysis and optimization techniques. However, as many modern real-time applications are increasingly distributed, the analysis and optimization of such systems have become significantly more complex. Distributed real-time systems involve multiple components spread across nodes, introducing challenges in communication, synchronization, and resource management. These systems often operate in heterogeneous environments, where performance, cost, and maintainability are critical concerns. The complexity is further compounded by the introduction of middleware layers that enable communication and coordination between different components, often abstracting away low-level details of data transfer and system integration. While middlewares can significantly simplify the development of distributed systems by enabling developers to focus on higher-level application logic, it also introduces new challenges related to ensuring real-time performance and predictability. Indeed, the interaction between middleware, operating systems (OS), and application software can have a profound impact on the timing behavior of real-time systems. As such, the correct configuration of middleware becomes essential in achieving the desired performance in time-sensitive applications. In this scenario, the Data Distribution Service (DDS), a middleware standard built on the publish-subscribe communication paradigm, has become a cornerstone in enabling efficient and scalable communication for distributed applications. It is widely adopted in critical domains such as smart cities, Industry 4.0, and autonomous driving, with automotive frameworks like ROS2, AUTOSAR Classic and Adaptive, and Autoware relying on DDS for data exchange. Despite its importance, DDS's multithreaded architecture and internal message dispatch policies complicate the enforcement of strict timing constraints required by real-time systems. To address these challenges, this thesis presents a comprehensive approach encompassing modeling, analysis, and optimization of DDS-based real-time distributed systems. First, a formal and abstract model of DDS is developed that is generalizable across any DDS-compliant implementation. The model captures both synchronous and asynchronous communication modes, and accounts for various internal scheduling policies and message dispatch mechanisms. A specific instance of this model is applied to eProsima’s FastDDS, a widely used DDS implementation, to conduct a detailed response-time analysis. This analysis provides concrete upper bounds on data-delivery latency and end-to-end response times, essential for ensuring the reliability and predictability of DDS-enabled systems in real-time environments. Additionally, an integration of the DDS analysis with executor-based scheduling analysis for ROS2-based systems is introduced, enabling a holistic perspective that covers both the DDS internals and ROS2-level scheduling. Furthermore, the thesis proposes a suite of analysis-driven optimization techniques aimed at enhancing the performance of DDS-based real-time systems. These techniques target the optimization of critical parameters, including classical thread priorities and DDS-specific parameters such as message dispatch policies, to ensure that timing constraints are met while minimizing latency. To facilitate the practical adoption of these methods, a novel Eclipse-based configuration framework is presented. This tool integrates the analysis and optimization techniques into a user-friendly interface, automating much of the system configuration process and reducing the complexity required for tuning DDS parameters. The validity of the proposed analysis and optimization techniques is demonstrated through extensive experimental evaluation on real-world platforms. Testbeds include the WATERS 2019 Industrial Challenge by Bosch, a representative automotive application, and the Autoware Reference System, a widely used framework in autonomous driving. Results show that the techniques significantly improve the predictability and efficiency of DDS-based systems, ensuring suitability for safety-critical real-time applications where timing guarantees are paramount. Through this work, system designers are enabled in the formal understanding of DDS’s real-time behavior, providing tools and methods for the analysis and systematic optimization of distributed systems. These contributions pave the way for more robust and time-predictable deployments of autonomous and industrial systems, meeting the demands of modern interconnected applications.