Moving from the traditional federated design paradigm, integration of different multi-criticality functionalities onto common computing platforms is increasingly being adopted by avionics industry, driven primarily by cost and related concerns. The design of such mixed-criticality real-time systems has been recognized to be a very important but challenging problem given that safety-critical functionalities with different criticality levels must be certified correct by statutory certification authorities (CAs) at varying levels of rigorousness. Traditionally, these certification requirements are addressed by assuming the highest criticality level for all the functionalities, which is rather pessimistic, however, since certifying at the highest criticality level implies the highest degree of assurance regarding the correct behavior of all the functionalities which need to be guaranteed by over-provisioning the required resources. Therefore, it is necessary to develop new design and analysis techniques that are able to facilitate the cerfication process while efficiently utilizing the computing resources. In this paper, we focus on scheduling and schedulability analysis problem for certifiable mixed-criticality real-time system on a uniprocessor platform. Firstly the widely-used traditional job model is generalized to mixed-criticality scenario where multiple different WCET values are specified for each job to reflect temporal constraints at all different levels of assurance. Then further investigation is conducted into an existing mixed-criticality scheduling strategy called Criticality Based Earliest Deadline First, the basic idea of which is to reserve time intervals for higher criticality jobs offline, while scheduling lower criticality jobs online at run-time using the so-called free slacks. However, the corresponding schedulability condition turns out to be incorrect. On the basis of this observation, an improved deadline-based dynamic mixed-criticality scheduling algorithm is proposed considering the design-for-certification issue, along with a revised sufficient schedulability condition deduced. And experiments conducted on randomly generated instances of jobs indicate that the proposed approach is effective and efficient.
展开▼
