Proposes a framework to describe and handle distributed systems as an interaction graph of elementary components. Components are discrete event systems operating on several state variables, and defining local dynamics on these variables. Components are interconnected by sharing variables, which defines the interaction graph of the compound system. They evolve asynchronously, with their own clock, so there is no notion of global time. This behavior is captured by the so-called true concurrency semantics on trajectories of the system. Just like the global system factorizes as a product of components, we prove that its trajectories also "factorize." As a consequence, the global system can be handled by parts, for example for state estimation; the global state of the system is never computed. This is a key to deal with large systems. This framework has been applied to design distributed diagnosis algorithms for telecommunication networks.