System oriented extensions to dataflow
Steve P. Landry · 1981
This dissertation addresses the extension of traditional dataflow modeling to include specifications to allow additional control over the execution environment. Three major studies are presented: (1) a comparative analysis of several proposed and existing dataflow models and architectures, (2) the specification of several extensions for generalizing traditional abstract dataflow models and providing the opportunity to express greater parallelism, and (3) the design and implementation of an evolutionary test bed for realizing the simulation of dataflow programs and systems. The extensions to the abstract dataflow model include: (1) a generalized firing rule to eliminate unnecessary synchronization introduced by requiring all inputs to be available before enabling nodal execution, (2) a mechanism for obtaining a higher degree of parallelism through replication of nodes, (3) a concept of generalized termination detection and signaling useful in supporting replication and streaming, and (4) the description of methods for supporting shared data objects and interprocess communications in a multiple process dataflow environment. The dataflow simulator, DFSS, was developed in support of this work and is a highly generalized facility for realizing the execution, debugging, and metering of dataflow programs.