ALTERNATIVE DATAFLOW MODEL

摘要

Dataflow architectures showed considerable potential however the initial interest has to some extent petered out. Difficulties arose with the cost of implementation, appropriate programming languages and operating systems, and ability to sustain development. We argue that there is a fundamental problem with dataflow relating to how a program is represented. This article explores representing a program by an unlimited acyclic network rather than a fixed one with either cycles or reuse. The state is captured in the edges rather than the nodes. We show how this model can address many of the shortcomings of dataflow and enable dataflow to realize its expected potential. The model shows promise in: improving parallelism and ease of implementation.