Distributed Implementation of SIGNAL: Scheduling & Graph Clustering

International audienceThis paper introduces the scheduling strategy and some key tools which have been designed for the distributed implementation of Signal, a real-time synchronous dataflow language. First, we motivate a scheduling strategy with respect to the reactivity and time-predictability requirements bound to real-time computing. Then, several key tools to implement this scheduling strategy are described. These tools are acting on the concept of Synchronous-Flow Dependence Graph (SFD Graph) which defines a generalization of Directed Acyclic Graph and constitutes the abstract representation of Signal programs. The tools presented in this paper are: (a) the abstraction of SFD graphs which enables grain-size tuning according to the target architecture, (b) the notion of scheduling over SFD graphs and (c) qualitative clustering tools based on the notion of Compositional Deadlock Consistency