DATAFLOW COMPUTING

A substitute to the Von Neumann model of calculation is dataflow computation model. In the dataflow model, control is tied to flow of data. Order of instructions in the program not plays any role on execution order. Execution of an instruction may take place when all the data needed by an instruction are available. Data is in continuing flow independent of reusable memory cells in addition its availability begins execution. Because, data is obtainable for various instructions at same time, these instructions are able to be executed in parallel. 

For the reason of using parallelism in calculation, Data Flow Graph notation is used to signify computations. In data flow graph, nodes represent commands of the program as well as edges represent data dependency among instructions.  As an example, the dataflow graph for instruction z = w × (x + y) is shown in Figure below

Figure: DFG for z = w × (x+y)

Data moves on edges of the graph in form of data tokens that comprise data values and status information. The asynchronous parallel calculation is determined by the firing rule that is expressed by means of tokens: a node of DFG can fire if there is a token on every of its input edges. If node fires then it consumes the input tokens, executes the connected operation and places result tokens on output edge. Graph nodes may be single instructions or tasks comprising multiple instructions. The benefit of the dataflow concept is that nodes of DFG may be self scheduled. Though, the hardware maintains to identify the availability of important data is much more complicated than von Neumann model. The Illustration of dataflow computer comprises MIT Tagged Token Data Flow architecture and Manchester Data Flow Machine.