Modeling and Analyzing Dataflow Applications on NoC-Based Many-Core Architectures

The advent of chip-level parallel architectures prompted a renewal of interest into dataflow process networks. The trend is to model an application independently from the architecture, then the model is morphed to best fit the target architecture. One downplayed aspect is the mapping of communications through the on-chip topology. The cost of such communications is often prevalent with regard to computations. This article establishes a dataflow process network called K-periodically Routed Graph (KRG), which serves the role of representing the various routing decisions during the transformation of a genuine application into a architecture-aware version for this application.