A Runtime Mechanism for Detection of Artificial Deadlocks in Process Networks

Kahn process network (KPN) is a popular model of computation for describing streaming applications arising in media and signal processing. KPN is a collection of sequential processes that communicate through unidirectional unbounded FIFOs. The illusion of unbounded FIFOs has to be met with finite memory in real implementations. This could potentially lead to a violation of Kahn semantics, thereby resulting in "artificial deadlocks". In this paper, we address this issue and present a runtime mechanism for early detection and resolution of such artificial deadlocks