Section: New Results

Communicating Networks of Data-Flow (sub)networks with limited memory

Participant : Robert de Simone.

Process Networks have been proposed a long time ago as models of concurrent, embedded streaming computations and communications, both amenable to formal analysis as models and executable as parallel program abstractions. As part of a larger effort at identifying precise connections between these models, programming models, and embedded parallel architectures altogether, we worked this year on the following problem: given a network of concurrent processes (Kahn-style) where each process is in turn a data-flow process network (SDF-style), can we decide in an efficient fashion (not NP-hard) whether a given assignment of communications to bounded local memories is schedulable (in a way that two simultaneous communications cannot require more than the available memory). A technical report is in preparation.