Optimised Adaptation of Mixed-Criticality Systems with Periodic Tasks on Uniform Multiprocessors in Case of Faults

The standard real-time computing model is based on a notion of deadline that represents both, a design parameter and a critical latency. The tolerance-based real-time computing model (TRTCM) extends the standard real-time computing model to reflect good engineering practice, where a safety margin is added to the critical latency to obtain the corresponding design parameter. The key feature of TRTCM is to add a tolerance range, which in combination with criticality specifications for services allows for fault-tolerant mixed-criticality scheduling with smooth degradation of service utility in case of resource shortage. In this paper we study the applicability of TRTCM for periodic tasks with mixed criticality on uniform multiprocessor systems. To optimise the system adaptation in case of resource shortage we formulate utility maximisation based on TRTCM as an optimisation problem. We describe two implementation strategies and derive an optimisation problem to be solved by a constraint solver for each. We use a video processing application for a concrete evaluation. Our results show that TRTCM can provide better utility than standard mixed-criticality approaches where lower criticality services are skipped in case of resource shortage.