Data-driven analysis and design

Data-driven analysis and design

Accurate models are essential in model-driven system performance engineering to ensure the accuracy of analysis results. Data-driven analysis and design techniques enable model learning, model validation and model calibration. System operational data can be used for monitoring performance targets during system operation, for diagnosing unexpected performance degradations, for system updates, and for system (re-)design when developing new generations or variants of the system at hand. Selecting the right data to be collected and lightweight, non-intrusive system instrumentation are essential.

Data collection

Operational data may be used to improve system performance. It is essential to collect the right data, via lightweight, non-intrusive system instrumentation and timing measurements to minimize the impact on system performance. It is important to ensure timing accuracy (e.g., via clock synchronization) and to consider storage and bandwidth limitations that determine whether the required data can be collected real-time or not.

Model learning

Operational data can be related to models through model validation and calibration, model learning, or digital twinning, balancing knowledge-driven and data-driven design approaches.

Related publications
Additional information