Optimization of Systems With Nested Design Space

Abstract

Finding a good solution to an engineering problem under given constraints can be formulated as an optimization problem over a set of parameters. Existing (semi-)automatic methodologies for this task usually assume that these design parameters form a homogeneous set. The paper analyses the structure of design space and introduces the formal concept of existence-dependent parameters, which appear for instance in the optimization of system architectures. The corresponding nested optimization problems can be solved more efficiently considering the dependency relations, and will often require a mix of models and optimization algorithms. The paper proposes a completely model-driven integration methodology for such problems and a prototype tool combining several existing domain-specific tools. An application example from the area of distributed embedded systems design of hardware architecture and software mapping demonstrates the applicability.