Dynamic reallocation of processes and system dimensioning in fault-tolerant control systems

Real-time control systems need often to take into account fault-tolerance capabilities to guarantee correct behavior or at least, to provide gracefully-degradable performance. In problem of hardware dimensioning and the optimum allocation of the computation are discussed. The author considers the problem of the dynamic reallocation of the computation, so that a higher number of faults can be tolerated, possibly with degraded performance and functionalities at low costs. The computation is modeled by using concurrent communicating processes, while the hardware structure considers multiprocessor distributed systems. The stochastic evaluation of the software performance is concerned with the capabilities of dealing with external events within a given maximum time. The hardware dimensioning is optimized at the same time as the software allocation. Redundant hardware resources are introduced to take into account the additional requirements of the spare processes. Faults are overcome by dynamically redistributing the computation. A single optimization frame is defined to identify a globally-optimum solution