Architecture analysis and optimization of reconfigurable, complex systems

Today networked systems are very complex and difficult to design. This causes by high communication traffic between devices, ressource load, several functionalities, memory usage, execution time, type of architecture, topology, device count or energy consumption. Keeping this factors in mind a special problem is which functionality should be processed on which device (mapping functions to architectures). Optimal system design results in a huge design space, because of a lot of system variants. A model based virtual prototype describes a system on abstract level of detail, enabling analysis and simulation in early design steps to improve or correct the system design. Added optimization algorithms can solve the np-complete mapping problem by creating and estimating new system variants. This combination of performance estimation and design space exploration results in a reconfigured optimized system. The paper shows an approach, that explains the steps from abstract system description via simulation to optimization. Tabu search, simulated annealing and greedy algorithm variants reconfigure the initial system. A theoretical example clarifies the methodology.