In mixed-criticality systems, the same computing platform is used to execute a workload consisting of safety-critical tasks and non-safety-critical ones. Whereas the former class of tasks exposes stringent timing and performance requirements that have to be guaranteed in terms of schedulability under pessimistic assumptions, the latter class can tolerate a certain level of violation of such requirements. A classic example of such a system is an automotive computing system, in which the same platforms execute driving-related tasks (braking and engine control) and infotainment applications.

Another trend of the last decade in computing systems has been the extensive exploitation of thread parallelization to dramatically improve performance of applications on multicore platforms. When considering a mixed-criticality system, parallelization presents at the same time an opportunity, as discussed, but also an additional issue in assessing the schedulability of real-time tasks.

The reviewed paper is probably the first work addressing the problem of scheduling mixed-criticality workload consisting of real-time parallel tasks on a multicore platform. In particular, the proposed solution is a mixed-criticality federated scheduling algorithm able to accommodate the workload, not completely known at design-time, on the available processing units; in the case of unschedulable situations, it minimizes the number of non-safety-critical tasks violating the deadline in critical situations.

The paper presents a very rigorous problem formulation and solution. Nevertheless, in order to demonstrate the applicability of the proposed solution in real-life scenarios, the scheduling algorithm has been implemented in a Linux operating system supporting the execution of OpenMP parallel applications.

Even if the topic and proposed solution are interesting and highly relevant and make the work highly valuable, in my opinion, the paper presents two main weakness: (1) its excessive formalism in the presentation of the solution, which could have been avoided by introducing a running example or some additional figures and diagrams illustrating the introduced models and details of the solution, and (2) some lacks in the discussion of the performed experimental campaigns. For instance, it is not explained how simulations have been performed and why, on the real system, the authors used synthetic benchmarks instead of real-world applications.