Hill-climbing heuristics for optimal hardware dimensioning and software allocation in fault-tolerant distributed systems

An optimum distributed architecture with fault-tolerance capabilities for a given software application may be obtained by allowing allocation algorithms to evolve without any existing-hardware constraint. Distributed software partitioning and allocation is done using the simulated annealing optimization algorithm. To define the cost function used by the optimization algorithm, a model for interacting processes constituting the software application is presented. Tuning of algorithm parameters has been considered to assure convergence at a reasonable cost in terms of computation time.