Compensating Adaptive Mixed Criticality Scheduling.

The majority of prior academic research into mixed criticality systems assumes that if high-criticality tasks continue to execute beyond the execution time limits at which they would normally finish, then further workload due to low-criticality tasks may be dropped in order to ensure that the high-criticality tasks can still meet their deadlines. Industry, however, takes a different view of the importance of low-criticality tasks, with many practical systems unable to tolerate the abandonment of such tasks. In this paper, we address the challenge of supporting genuinely graceful degradation in mixed criticality systems, thus avoiding the abandonment problem. We explore the Compensating Adaptive Mixed Criticality (C-AMC) scheduling scheme. C-AMC ensures that both high- and low-criticality tasks meet their deadlines in both normal and degraded modes. Under C-AMC, jobs of low-criticality tasks, released in degraded mode, execute imprecise versions that provide essential functionality and outputs of sufficient quality, while also reducing the overall workload. This compensates, at least in part, for the overload due to the abnormal behavior of high-criticality tasks. C-AMC is based on fixed-priority preemptive scheduling and hence provides a viable migration path along which industry can make an evolutionary transition from current practice.