DIF - A Language for Dataflow Graph Specification and Exchange

Abstract

Abstract : A major impediment to broad acceptance of dataflow specifications for high performance applications is the absence of any standard for dataflow language. While virtually all dataflow graph based specification methods and supporting software tools are based on a common dataflow mathematical model, they are never the less all mutually incompatible. There are a number of dataflow tools and environments that either execute the dataflow specifications directly or automatically translate the specifications into source code for executable realizations of the specifications. Transferring dataflow graphs or reusing legacy designs and specifications from a different dataflow programming environment is not possible. In the absence of anything like ANSI language standards, developers are reluctant to utilize the promising dataflow technology because using a particular vendor's tools is a long term commitment to the vendor as well as the technology. The fact that dataflow tool vendors are for the most part small businesses only exacerbates the problem. The Dataflow Interchange Format (DIF) is an ongoing research project at the University of Maryland (UMD). Its objective is the development of a vendor neutral language for specification of dataflow graphs. Its acceptance as an industry standard by dataflow tool vendors would enable exchange of dataflow graphs among supporting vendors tools and programming environments. Developers could save and reuse legacy dataflow graphs from all vendors tools used in past developments. Government could provide dataflow graph specifications in DIF from other contracts as GFl with the expectation that they could be readily imported into whatever dataflow environment in use.