Data-flow to Von Neumann : the SIGNAL approach

In real-time applications design, two parts have to be considered : such an application handles values (or more precisely, sequences of values), as any information system ; moreover, (some notion of) time has an effect upon the behaviour of the application ; not only scheduling but also valued results may depend upon the instants of events occurrences. In the traditional approaches, the time is a Daemon which unforeseeable behaviour the designer has to cope with. Conversely, the synchronous approach allows to partly look time as other values : the time is nothing but a partial order of events on which computing is possible. Merging synchronous and data-flow principles, Signal has been defined from a small set of operators, for designing real-time programs running in (possibly) distributed environments. In this context, it is necessary to be able to generate a family of parallel processes (each process may be a sequential, a data-flow or a parallel proces). This may be achieved by partitioning the initial nodes of a synchronized data-flow graph (at a micro-data-flow level) to build a new (macro-data-flow) graph such that, firstly a sequential implementation may be locally calculated and secondly, this implementation is correct whith respect to the logical time properties of the whole graph. In this paper, after a definition of Signal kernel, we describe the graph and the synchronization space associated with each process, finally properties of graphs, with respect to different scheduling policies, are studied in the framework of signal-graphs (graphs with distinguished input and output nodes)