We are interested in mapping hard real-time applications on distributed heterogeneous architectures. An application is modeled as a set of tasks, and we consider a fixed-priority preemptive scheduling policy. We target the early design phases, when decisions have a high impact on the subsequent implementation choices. However, due to a lack of information, the early design phases are characterized by uncertainties, e.g., in the worst-case execution times (wcets), or in the functionality requirements. We model uncertainties in the wcets using the “percentile method”. The uncertainties in the functionality requirements are captured using “future scenarios”, which are task sets that model functionality likely to be added in the future. In this context, we derive a mapping of tasks in the application, such that the resulted implementation is both robust and flexible. Robust means that the application has a high chance of being schedulable, considering the wcet uncertainties, whereas a flexible mapping has a high chance to successfully accommodate the future scenarios. We propose a Genetic Algorithm-based approach to solve this optimization problem. Extensive experiments show the importance of taking into account the uncertainties during the early design phases.
|Title of host publication||DATE '12:Proceedings of the Conference on Design, Automation and Test in Europe|
|Publisher||Association for Computing Machinery|
|Publication status||Published - 2012|
|Event||2012 Design, Automation & Test in Europe Conference & Exhibition (DATE) - Dresden, Germany|
Duration: 12 Mar 2012 → 16 Mar 2012
|Conference||2012 Design, Automation & Test in Europe Conference & Exhibition (DATE)|
|Period||12/03/2012 → 16/03/2012|