Simulation Based Execution Time Analysis of SDF Applications on Heterogeneous MPSoCs Using Measured Delay Distributions

Kurzfassung

The complexity of tasks of modern embedded systems changed from simple control algorithms to image-processing or even Artificial Intelligence. During design optimization of Heterogeneous Multi-Processor Systems-on-a-Chip early performance analysis is required to guarantee a certain quality of service. The proposed simulation based approach to analyze timing behavior of software addresses accuracy by maintaining scalability. Hardware is modeled as a set of composable tiles. The software model differentiates between computation and communication phases. This allows to determine interference on shared resources. Measured delays are used for the performance model. By keeping the distribution of execution delay the analysis results are more representative to the actual behavior of the real system. The approach presented in this thesis allowed a prediction of the average execution time of a HW/SW configuration with an error less than 5%.