Towards an efficient simulation of multi-language descriptions of heterogeneous systems

Heterogeneous systems design requires executing models using different simulators. Usually, simulators are connected and synchronized together by using mechanisms such as shared memory or TCP/IP. Current approaches have many drawbacks: they are not adapted to distributed environments, the simulation performance may be very disappointing, and each simulator has its own paradigm. We propose a new approach to simulate models described with different languages such as SystemC or ESyS.NFT on a distributed environment. All the models are translated into a unique internal format in order to be able to optimize a distributed and multi-paradigm simulation of the models. Optimizations include different mechanisms such as regrouping of threads in a single one while respecting the semantics of the models or the splitting of a model in order to execute it on several computer resources.