Creating New Real-Time Scheduling Policies: Why and How?

In real-time systems, schedulability is mandatory but other application-dependent performance criteria, such as average response time jitters, are most generally of interest. This is typically the case for computer-controlled systems that are known to be very sensitive to various delays induced by resource sharing; the delays reduce the performances of the system or even may jeopardize the stability of the system. We first define the properties that a "good" real-time scheduling algorithm must possess. Then, we exhibit a class of easily-implementable policies that should be well suited to various applicative contexts because these policies provide good trade-off between feasibility and the satisfaction of the application-dependent criteria. We propose a schedulability analysis generic for all policies within this class and simulation models for evaluating other performance criteria.