Cyclic Executives, Multi-Core Platforms and Mixed Criticality Applications

Historically safety-critical real-time systems have been implemented using a cyclic executive (CE). Here a series of frames (minor cycles) are executed in sequence. Once the series is complete the sequence is repeated. The duration of the full sequence is often known as the major cycle. Within each frame, units of computation (jobs) are executed, again in sequence. Although there are a number of drawbacks to the use of CEs they have the advantage of being fully deterministic and efficiently implemented. For multi-core platforms, running a set of frames on each core is an obvious extension to the single core approach. Here there is advantage in coordinating the execution of the cores so that frames are released at the same time across all cores. For mixed criticality systems, the requirement for separation would imply that, at any time, code of the same criticality must execute on all cores. In this paper we consider how this requirement can be met and the performance, in terms of schedulability, it delivers. We consider partitioned and globally allocated work. For partitioned systems an allocation scheme is developed. For globally scheduled schemes we develop a polynomial-time sufficient schedulability test that determines whether a given mixed-criticality system is schedulable, and constructs a schedule if it is.