Performance and dependability assessment is necessary for the sound design of complex critical sys-tems. Traditional methods are data-driven or model-driven. Real-world data utilization in data-driven performance and reliability assessment focuses on evaluating system behavior under operational or experimental conditions. A key element is the integration of diverse data sources, including opera-tional data, logs and failure reports. Applying machine learning to reliability data makes it possible to detect anomalies, perform predictive maintenance, and improve fault tolerance in many systems, from cloud-native applications to autonomous systems and IoT networks.

Modeling approaches provide structured frameworks to capture and predict system behavior, enabling the identification of performance and reliability bottlenecks early in the design phase when no data is available at the system level. Model types in use include queueing networks and fault trees, reliability block diagrams as well as Markov chains, semi-Markov and Markov regenerative processes, and sto-chastic Petri nets. Models can be solved by discrete-event simulation or by analytic-numeric methods. When analytic-numerical methods become impractical due to system scale or irregularity, simulation offers the flexibility needed for performance and dependability analysis. However, it requires statistical estimation, like data-driven methods, and can involve long run times, particularly in reliability model-ing for rare events.

With increased complexity of systems, we believe that integrating data-driven and model-driven methods on the one hand and hybrid solution methods combining simulative with analytic-numeric solutions on the other hand will become necessary. While isolated attempts at such an integration do exist, this Dagstuhl Seminar will bring intense focus on issues arising in interfacing data-driven and model-driven methods on the one hand and simulative and analytic-numeric solution on the other hand. This will be done while keeping in mind real-world applications and software packages to sup-port such integration.

Emerging technologies offer significant opportunities for innovation in performance and dependability assessment. While every community develops its specific solutions, integrating models—combining data-driven and model-based (analytical, numerical, and simulation) approaches—allows system de-signers to optimize performance and reliability in ways that were not possible with traditional, silo-ed methods. By leveraging model-driven and data-driven assessments together, designers can test config-urations, identify bottlenecks, and predict system behavior under diverse conditions without directly affecting operational systems. This Dagstuhl Seminar aims to join researchers from the communities of data-driven dependability assessment, the analytic and analytic-numeric solution of performance and dependability models, and simulation-based solution of performance and dependability models. By finding and fostering interfaces between these fields, the seminar will act as a bridge for integrat-ing approaches, facilitating the integration of existing and emerging techniques to address real-world challenges in system evaluation. It will emphasize various evaluation methods, including data-driven, simulation, hybrid analytic-numeric techniques, and multi-level approaches. Additionally, tool en-hancements/demonstrations will be highlighted to showcase practical applications. Emerging areas such as AI-based systems modeling, digital twins, smart grids, and resilience in autonomous and IoT systems will provide real-world context.

The results of the seminar are planned to be submitted to either IEEE Computer magazine or IEEE Reliability magazine. Ideally, the discussions during the seminar will lead to a collaborative book compiled from chapters written by the participants.

Creative Commons BY 4.0

Paulo Romero Martins Maciel, Anne Remke, Kishor S. Trivedi, and Armin Zimmermann