Scheduling mixed-criticality multi-core systems to maximise resource utilisation

Abstract

Current practice in the safety-critical industry is to run critical application programs on dedicated processors. Adopting widely available multi-core processors to run programs of different levels of criticality in parallel on the same chip would allow the required performance to be achieved with lower energy and material costs. However, the inherent run-time variability would be exacerbated by indirect interference through shared resources, to the extent that no overall benefit would be derived from multi-core technology. We consider the case of a dual-core ARM processor with a real-time program assigned to one core and a best-effort program assigned to the other. An existing technique from the literature allows us to guarantee safety by stopping the best-effort program – and therefore the interference – whenever a real-time deadline is in danger of being overrun. We show that combining this technique with quality-of-service strategies can result in significant improvements in processor utilisation while maintaining the safety guarantee.