Three Search Strategies for Architecture Synthesis and Partitioning of Real-Time Systems

This report studies the problem of automatically selecting a suitable system architecture for implementing a real-time application. Given a library of hardware components, it is shown how an architecture can be synthesized with the goal of fulfilling the real-time constraints stated in the system's specification. In case the selected architecture contains several processing units, the specification is partitioned by assigning tasks to processing units. We investigate the use of three meta-heuristic search algorithms to solve the problem: genetic algorithms, simulated annealing, and tabu search; and it is described in detail how these can be adapted to the architecture synthesis problem. Their relative merits are discussed at length, as is the importance of scheduling to the solution quality. This work has been supported by the Swedish National Board for Industrial and Technical Development (NUTEK). Contents 1 Introduction 1 2 Related work 1 3 Problem description 3 3.1 Behavioural..