SATURNE: a reactive-anytime programming model for intelligent embedded real-time systems

One of the major challenge of next embedded systems is to involve intelligence while preserving the classical real-time properties: (a) reactivity, i.e. the capability to react continuously towards asynchronous inputs, and (b) predictability. Furthermore, one of the main property which is requested in this kind of systems operating in a highly non-deterministic outside environment, is adaptability to new and unexpected conditions, and particularly to dynamic temporal deadlines. Strong predictability often means that it should be possible to prove both the functional and temporal behavior of a system, in response to any combination of signals from outside environment, and particularly to prove that critical tasks meet their deadlines. On the contrary, in non-deterministic environments involving dynamic timing constraints, the assumption that critical tasks meet their deadlines can not be guaranteed. In that sense, strong predictability and adaptability are contradictory requirements. Furthermore, intelligence often means complex computations. In that sense reactivity and punctuality seem also to be incompatible with reasoning capabilities. The aim of this paper is to investigate a programming model, called SATURNE, addressing the issue of intelligent real-time systems, that is how to mix adaptability, reactivity and predictability. We only focuses on adaptability to temporal deadlines, addressing the problem of guaranteeing a response when temporal deadlines may be statically unpredictable. SATURNE is based on a mixed reactive-anytime approach. The main idea is to introduce firstly a reactive model of computation providing predictability, and secondly an anytime model of computation providing adaptability to dynamic temporal deadlines